From e8dc67fcf15d83364d969f586c1cd03f225542e3 Mon Sep 17 00:00:00 2001 From: xavidop Date: Mon, 6 Jul 2026 13:58:14 +0200 Subject: [PATCH 1/6] feat: more plugins --- README.md | 4 +- docs/astro.config.mjs | 3 + .../src/content/docs/agents/session-stores.md | 100 ++- docs/src/content/docs/getting-started.mdx | 2 +- docs/src/content/docs/index.mdx | 33 + docs/src/content/docs/models.mdx | 2 +- docs/src/content/docs/plugins/chroma.md | 66 ++ .../docs/plugins/firebase-functions.md | 2 +- .../docs/plugins/firebase-vector-store.md | 8 +- docs/src/content/docs/plugins/google-genai.md | 4 +- docs/src/content/docs/plugins/mongodb.md | 83 ++ .../src/content/docs/plugins/opentelemetry.md | 2 +- docs/src/content/docs/plugins/postgresql.md | 15 + docs/src/content/docs/plugins/qdrant.md | 70 ++ .../main/java/com/google/genkit/Genkit.java | 2 +- plugins/chroma/pom.xml | 78 ++ .../chroma/ChromaCollectionConfig.java | 206 +++++ .../genkit/plugins/chroma/ChromaPlugin.java | 201 +++++ .../plugins/chroma/ChromaVectorStore.java | 311 +++++++ .../genkit/plugins/chroma/package-info.java | 25 + .../chroma/ChromaCollectionConfigTest.java | 70 ++ .../plugins/chroma/ChromaPluginTest.java | 55 ++ .../plugins/chroma/ChromaVectorStoreTest.java | 111 +++ plugins/evaluators/README.md | 10 +- .../plugins/evaluators/EvaluatorsPlugin.java | 2 +- .../EvaluatorsPluginOptionsTest.java | 4 +- plugins/firebase/README.md | 8 +- .../plugins/firebase/FirebasePlugin.java | 2 +- .../retriever/FirestoreRetrieverConfig.java | 4 +- .../plugins/googlegenai/GeminiModel.java | 2 +- plugins/mongodb/pom.xml | 80 ++ .../genkit/plugins/mongodb/MongoPlugin.java | 183 +++++ .../plugins/mongodb/MongoVectorStore.java | 363 +++++++++ .../mongodb/MongoVectorStoreConfig.java | 355 ++++++++ .../genkit/plugins/mongodb/package-info.java | 26 + .../mongodb/session/MongoSessionStore.java | 723 +++++++++++++++++ .../session/MongoSessionStoreOptions.java | 249 ++++++ .../plugins/mongodb/session/package-info.java | 28 + .../plugins/mongodb/MongoPluginTest.java | 65 ++ .../mongodb/MongoVectorStoreConfigTest.java | 94 +++ .../plugins/mongodb/MongoVectorStoreTest.java | 138 ++++ .../session/MongoSessionStoreOptionsTest.java | 78 ++ .../session/MongoSessionStoreTest.java | 202 +++++ plugins/pinecone/README.md | 10 +- .../plugins/pinecone/PineconeIndexConfig.java | 2 +- .../plugins/pinecone/PineconePlugin.java | 2 +- .../genkit/plugins/pinecone/package-info.java | 2 +- plugins/postgresql/README.md | 4 +- .../plugins/postgresql/PostgresPlugin.java | 2 +- .../postgresql/PostgresTableConfig.java | 2 +- .../plugins/postgresql/package-info.java | 2 +- .../session/PostgresSessionStore.java | 761 ++++++++++++++++++ .../session/PostgresSessionStoreOptions.java | 243 ++++++ .../postgresql/session/package-info.java | 27 + .../PostgresSessionStoreOptionsTest.java | 76 ++ .../session/PostgresSessionStoreTest.java | 212 +++++ plugins/qdrant/pom.xml | 78 ++ .../qdrant/QdrantCollectionConfig.java | 258 ++++++ .../genkit/plugins/qdrant/QdrantPlugin.java | 176 ++++ .../plugins/qdrant/QdrantVectorStore.java | 333 ++++++++ .../genkit/plugins/qdrant/package-info.java | 25 + .../qdrant/QdrantCollectionConfigTest.java | 78 ++ .../plugins/qdrant/QdrantPluginTest.java | 48 ++ .../plugins/qdrant/QdrantVectorStoreTest.java | 110 +++ plugins/weaviate/README.md | 6 +- .../weaviate/WeaviateCollectionConfig.java | 4 +- .../plugins/weaviate/WeaviatePlugin.java | 2 +- .../genkit/plugins/weaviate/package-info.java | 2 +- pom.xml | 8 + samples/README.md | 5 + samples/agents-mongo-session/README.md | 65 ++ samples/agents-mongo-session/pom.xml | 87 ++ samples/agents-mongo-session/run.sh | 7 + .../genkit/samples/MongoSessionAgentApp.java | 119 +++ .../src/main/resources/logback.xml | 24 + samples/agents-postgres-session/README.md | 70 ++ samples/agents-postgres-session/pom.xml | 87 ++ samples/agents-postgres-session/run.sh | 7 + .../samples/PostgresSessionAgentApp.java | 135 ++++ .../src/main/resources/logback.xml | 24 + samples/chroma/README.md | 60 ++ samples/chroma/pom.xml | 86 ++ samples/chroma/run.sh | 7 + .../samples/chroma/ChromaRAGSample.java | 163 ++++ samples/chroma/src/main/resources/logback.xml | 20 + samples/firebase/README.md | 2 +- .../samples/firebase/FirestoreRAGSample.java | 4 +- .../functions/GeneratePoemFunction.java | 2 +- samples/google-genai/README.md | 6 +- .../google/genkit/samples/GoogleGenAIApp.java | 6 +- samples/mongo-vector/README.md | 65 ++ samples/mongo-vector/pom.xml | 86 ++ samples/mongo-vector/run.sh | 7 + .../samples/mongo/MongoVectorRAGSample.java | 171 ++++ .../src/main/resources/logback.xml | 21 + samples/pinecone/README.md | 6 +- .../samples/pinecone/PineconeRAGSample.java | 4 +- samples/postgresql/README.md | 4 +- .../samples/postgresql/PostgresRAGSample.java | 4 +- samples/qdrant/README.md | 63 ++ samples/qdrant/pom.xml | 86 ++ samples/qdrant/run.sh | 7 + .../samples/qdrant/QdrantRAGSample.java | 166 ++++ samples/qdrant/src/main/resources/logback.xml | 20 + samples/weaviate/README.md | 4 +- .../samples/weaviate/WeaviateRAGSample.java | 4 +- .../SKILL.md | 12 +- 107 files changed, 8178 insertions(+), 80 deletions(-) create mode 100644 docs/src/content/docs/plugins/chroma.md create mode 100644 docs/src/content/docs/plugins/mongodb.md create mode 100644 docs/src/content/docs/plugins/qdrant.md create mode 100644 plugins/chroma/pom.xml create mode 100644 plugins/chroma/src/main/java/com/google/genkit/plugins/chroma/ChromaCollectionConfig.java create mode 100644 plugins/chroma/src/main/java/com/google/genkit/plugins/chroma/ChromaPlugin.java create mode 100644 plugins/chroma/src/main/java/com/google/genkit/plugins/chroma/ChromaVectorStore.java create mode 100644 plugins/chroma/src/main/java/com/google/genkit/plugins/chroma/package-info.java create mode 100644 plugins/chroma/src/test/java/com/google/genkit/plugins/chroma/ChromaCollectionConfigTest.java create mode 100644 plugins/chroma/src/test/java/com/google/genkit/plugins/chroma/ChromaPluginTest.java create mode 100644 plugins/chroma/src/test/java/com/google/genkit/plugins/chroma/ChromaVectorStoreTest.java create mode 100644 plugins/mongodb/pom.xml create mode 100644 plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/MongoPlugin.java create mode 100644 plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/MongoVectorStore.java create mode 100644 plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/MongoVectorStoreConfig.java create mode 100644 plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/package-info.java create mode 100644 plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/session/MongoSessionStore.java create mode 100644 plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/session/MongoSessionStoreOptions.java create mode 100644 plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/session/package-info.java create mode 100644 plugins/mongodb/src/test/java/com/google/genkit/plugins/mongodb/MongoPluginTest.java create mode 100644 plugins/mongodb/src/test/java/com/google/genkit/plugins/mongodb/MongoVectorStoreConfigTest.java create mode 100644 plugins/mongodb/src/test/java/com/google/genkit/plugins/mongodb/MongoVectorStoreTest.java create mode 100644 plugins/mongodb/src/test/java/com/google/genkit/plugins/mongodb/session/MongoSessionStoreOptionsTest.java create mode 100644 plugins/mongodb/src/test/java/com/google/genkit/plugins/mongodb/session/MongoSessionStoreTest.java create mode 100644 plugins/postgresql/src/main/java/com/google/genkit/plugins/postgresql/session/PostgresSessionStore.java create mode 100644 plugins/postgresql/src/main/java/com/google/genkit/plugins/postgresql/session/PostgresSessionStoreOptions.java create mode 100644 plugins/postgresql/src/main/java/com/google/genkit/plugins/postgresql/session/package-info.java create mode 100644 plugins/postgresql/src/test/java/com/google/genkit/plugins/postgresql/session/PostgresSessionStoreOptionsTest.java create mode 100644 plugins/postgresql/src/test/java/com/google/genkit/plugins/postgresql/session/PostgresSessionStoreTest.java create mode 100644 plugins/qdrant/pom.xml create mode 100644 plugins/qdrant/src/main/java/com/google/genkit/plugins/qdrant/QdrantCollectionConfig.java create mode 100644 plugins/qdrant/src/main/java/com/google/genkit/plugins/qdrant/QdrantPlugin.java create mode 100644 plugins/qdrant/src/main/java/com/google/genkit/plugins/qdrant/QdrantVectorStore.java create mode 100644 plugins/qdrant/src/main/java/com/google/genkit/plugins/qdrant/package-info.java create mode 100644 plugins/qdrant/src/test/java/com/google/genkit/plugins/qdrant/QdrantCollectionConfigTest.java create mode 100644 plugins/qdrant/src/test/java/com/google/genkit/plugins/qdrant/QdrantPluginTest.java create mode 100644 plugins/qdrant/src/test/java/com/google/genkit/plugins/qdrant/QdrantVectorStoreTest.java create mode 100644 samples/agents-mongo-session/README.md create mode 100644 samples/agents-mongo-session/pom.xml create mode 100755 samples/agents-mongo-session/run.sh create mode 100644 samples/agents-mongo-session/src/main/java/com/google/genkit/samples/MongoSessionAgentApp.java create mode 100644 samples/agents-mongo-session/src/main/resources/logback.xml create mode 100644 samples/agents-postgres-session/README.md create mode 100644 samples/agents-postgres-session/pom.xml create mode 100755 samples/agents-postgres-session/run.sh create mode 100644 samples/agents-postgres-session/src/main/java/com/google/genkit/samples/PostgresSessionAgentApp.java create mode 100644 samples/agents-postgres-session/src/main/resources/logback.xml create mode 100644 samples/chroma/README.md create mode 100644 samples/chroma/pom.xml create mode 100755 samples/chroma/run.sh create mode 100644 samples/chroma/src/main/java/com/google/genkit/samples/chroma/ChromaRAGSample.java create mode 100644 samples/chroma/src/main/resources/logback.xml create mode 100644 samples/mongo-vector/README.md create mode 100644 samples/mongo-vector/pom.xml create mode 100755 samples/mongo-vector/run.sh create mode 100644 samples/mongo-vector/src/main/java/com/google/genkit/samples/mongo/MongoVectorRAGSample.java create mode 100644 samples/mongo-vector/src/main/resources/logback.xml create mode 100644 samples/qdrant/README.md create mode 100644 samples/qdrant/pom.xml create mode 100755 samples/qdrant/run.sh create mode 100644 samples/qdrant/src/main/java/com/google/genkit/samples/qdrant/QdrantRAGSample.java create mode 100644 samples/qdrant/src/main/resources/logback.xml diff --git a/README.md b/README.md index 85b180b9d..d222f4b92 100644 --- a/README.md +++ b/README.md @@ -438,7 +438,7 @@ Genkit genkit = Genkit.builder() .addRetriever(FirestoreRetrieverConfig.builder() .name("my-docs") .collection("documents") - .embedderName("googleai/text-embedding-004") + .embedderName("googleai/gemini-embedding-001") .vectorField("embedding") .contentField("content") .distanceMeasure(FirestoreRetrieverConfig.DistanceMeasure.COSINE) @@ -476,7 +476,7 @@ public class MyFunction implements HttpFunction { genkit.defineFlow("generatePoem", String.class, String.class, (ctx, topic) -> { return genkit.generate(GenerateOptions.builder() - .model("googleai/gemini-2.0-flash") + .model("googleai/gemini-2.5-flash") .prompt("Write a poem about: " + topic) .build()).getText(); }); diff --git a/docs/astro.config.mjs b/docs/astro.config.mjs index e3ed7269a..9bae49eff 100644 --- a/docs/astro.config.mjs +++ b/docs/astro.config.mjs @@ -142,6 +142,9 @@ export default defineConfig({ { label: "Weaviate", slug: "plugins/weaviate" }, { label: "PostgreSQL (pgvector)", slug: "plugins/postgresql" }, { label: "Pinecone", slug: "plugins/pinecone" }, + { label: "Chroma", slug: "plugins/chroma" }, + { label: "Qdrant", slug: "plugins/qdrant" }, + { label: "MongoDB", slug: "plugins/mongodb" }, ], }, { diff --git a/docs/src/content/docs/agents/session-stores.md b/docs/src/content/docs/agents/session-stores.md index a8b721e3f..4ecf3d3fc 100644 --- a/docs/src/content/docs/agents/session-stores.md +++ b/docs/src/content/docs/agents/session-stores.md @@ -1,6 +1,6 @@ --- title: Session Stores -description: Persist agent session state with in-memory, file-based, Firestore, DynamoDB, or Cosmos DB session stores. +description: Persist agent session state with in-memory, file-based, Firestore, DynamoDB, Cosmos DB, PostgreSQL, or MongoDB session stores. --- A `SessionStore` is where a server-managed agent keeps its sessions. Pass one to `.store(...)` when you define an agent to persist snapshots; omit it to run the agent client-managed (stateless). Several implementations ship out of the box, and you can write your own. @@ -12,8 +12,10 @@ A `SessionStore` is where a server-managed agent keeps its sessions. Pass one | `FirestoreSessionStore` | Google Cloud Firestore | Production on Google Cloud (Cloud Run, Firebase Functions) | | `DynamoDbSessionStore` | Amazon DynamoDB | Production on AWS | | `CosmosSessionStore` | Azure Cosmos DB | Production on Azure | +| `PostgresSessionStore` | PostgreSQL | Production on any SQL-first stack | +| `MongoSessionStore` | MongoDB | Production on any document-first stack | -If you plan to use `chat.abort()` or the `/abort` HTTP endpoint, pick a store that supports change notifications: `FileSessionStore`, `FirestoreSessionStore`, `DynamoDbSessionStore`, and `CosmosSessionStore` do; `InMemorySessionStore` does not, so `abort()` is a no-op there. See [Sessions](../sessions#aborting-a-turn). +If you plan to use `chat.abort()` or the `/abort` HTTP endpoint, pick a store that supports change notifications: `FileSessionStore`, `FirestoreSessionStore`, `DynamoDbSessionStore`, `CosmosSessionStore`, `PostgresSessionStore`, and `MongoSessionStore` do; `InMemorySessionStore` does not, so `abort()` is a no-op there. See [Sessions](../sessions#aborting-a-turn). ## InMemorySessionStore @@ -199,6 +201,100 @@ CosmosSessionStore> store = The default shard size is 1 MiB, kept under Cosmos DB's 2 MB document-size limit. `CosmosSessionStore` supports `onSnapshotStateChange` (via polling), so `chat.abort()` works. +## PostgresSessionStore + +Stores snapshots in PostgreSQL. It lives in the PostgreSQL plugin, so add the dependency: + +```xml + + com.google.genkit + genkit-plugin-postgresql + ${genkit.version} + +``` + +```java +import com.google.genkit.plugins.postgresql.session.PostgresSessionStore; +import com.google.genkit.plugins.postgresql.session.PostgresSessionStoreOptions; +import org.postgresql.ds.PGSimpleDataSource; + +PGSimpleDataSource dataSource = new PGSimpleDataSource(); +dataSource.setUrl("jdbc:postgresql://localhost:5432/genkit"); +dataSource.setUser("postgres"); +dataSource.setPassword("postgres"); + +PostgresSessionStore> store = + new PostgresSessionStore<>(dataSource); // uses the default "genkit_sessions" table + +Agent> agent = genkit.beta().defineAgent( + AgentConfig.>builder() + .name("myAgent") + .system("You are a helpful assistant.") + .store(store) + .build()); +``` + +All records live in a single table keyed by `(pk, id)` with a JSONB `doc` payload and a `version` column for optimistic concurrency. Create the table ahead of time, or let the store create it on first use: + +```java +PostgresSessionStore> store = + new PostgresSessionStore<>( + dataSource, + PostgresSessionStoreOptions.builder() + .tableName("genkit_sessions") // table name (default "genkit_sessions") + .createTableIfNotExists(true) // auto-create the table on first use (default false) + .pollIntervalMs(500) // how often change subscribers are polled (default 2000) + .build()); +``` + +The default shard size is 1 MiB. `PostgresSessionStore` supports `onSnapshotStateChange` (via polling), so `chat.abort()` works. + +## MongoSessionStore + +Stores snapshots in MongoDB. It lives in the MongoDB plugin, so add the dependency: + +```xml + + com.google.genkit + genkit-plugin-mongodb + ${genkit.version} + +``` + +```java +import com.google.genkit.plugins.mongodb.session.MongoSessionStore; +import com.google.genkit.plugins.mongodb.session.MongoSessionStoreOptions; +import com.mongodb.client.MongoClient; +import com.mongodb.client.MongoClients; + +MongoClient client = MongoClients.create("mongodb://localhost:27017"); + +MongoSessionStore> store = + new MongoSessionStore<>(client); // uses database "genkit", collection "genkit_sessions" + +Agent> agent = genkit.beta().defineAgent( + AgentConfig.>builder() + .name("myAgent") + .system("You are a helpful assistant.") + .store(store) + .build()); +``` + +All records live in a single collection whose `_id` is `::`; each document carries a `version` field for optimistic concurrency. The database and collection are created automatically on first write. Tune it with `MongoSessionStoreOptions`: + +```java +MongoSessionStore> store = + new MongoSessionStore<>( + client, + MongoSessionStoreOptions.builder() + .databaseName("genkit") // database name (default "genkit") + .collectionName("genkit_sessions") // collection name (default "genkit_sessions") + .pollIntervalMs(500) // how often change subscribers are polled (default 2000) + .build()); +``` + +The default shard size is 1 MiB, kept under MongoDB's 16 MB document-size limit. `MongoSessionStore` supports `onSnapshotStateChange` (via polling), so `chat.abort()` works. + ## Writing your own store Implement `SessionStore` to use any backend — a database, a cache, a cloud object store: diff --git a/docs/src/content/docs/getting-started.mdx b/docs/src/content/docs/getting-started.mdx index 77f91b992..a8b8de133 100644 --- a/docs/src/content/docs/getting-started.mdx +++ b/docs/src/content/docs/getting-started.mdx @@ -186,7 +186,7 @@ Genkit genkit = Genkit.builder() ModelResponse response = genkit.generate( GenerateOptions.builder() - .model("googleai/gemini-2.0-flash") + .model("googleai/gemini-2.5-flash") .prompt("Tell me a fun fact!") .build()); ``` diff --git a/docs/src/content/docs/index.mdx b/docs/src/content/docs/index.mdx index f8ad63954..76d4f5336 100644 --- a/docs/src/content/docs/index.mdx +++ b/docs/src/content/docs/index.mdx @@ -214,6 +214,36 @@ The fastest way to build AI features into your Java apps. +## Build stateful agents + +Define server-managed agents with memory, tools, and multi-agent delegation. Persist conversations in the store of your choice. + +```java +import com.google.genkit.agent.AgentConfig; +import com.google.genkit.ai.agent.*; + +Agent> assistant = genkit.beta().defineAgent( + AgentConfig.>builder() + .name("assistant") + .system("You are a helpful assistant. Remember what the user tells you.") + .model("googleai/gemini-3-flash") + .store(sessionStore) // persist sessions in Firestore, DynamoDB, Cosmos DB, PostgreSQL, MongoDB, ... + .build()); + +AgentChat> chat = assistant.chat(); +chat.send("My name is Ada Lovelace."); +String reply = chat.send("What is my name?").text(); // "Your name is Ada Lovelace." +``` + + + + + + + + + + ## Supported Providers @@ -238,6 +268,9 @@ The fastest way to build AI features into your Java apps. + + + diff --git a/docs/src/content/docs/models.mdx b/docs/src/content/docs/models.mdx index 0c9fd1c9f..d46d760d3 100644 --- a/docs/src/content/docs/models.mdx +++ b/docs/src/content/docs/models.mdx @@ -59,7 +59,7 @@ genkit.generate(GenerateOptions.builder() // Use Gemini genkit.generate(GenerateOptions.builder() - .model("googleai/gemini-2.0-flash") + .model("googleai/gemini-2.5-flash") .prompt("Hello!").build()); // Use Claude diff --git a/docs/src/content/docs/plugins/chroma.md b/docs/src/content/docs/plugins/chroma.md new file mode 100644 index 000000000..ae44a5422 --- /dev/null +++ b/docs/src/content/docs/plugins/chroma.md @@ -0,0 +1,66 @@ +--- +title: Chroma +description: Chroma vector database integration for Genkit RAG workflows. +--- + +The Chroma plugin registers retrievers and indexers backed by a [Chroma](https://www.trychroma.com/) server (v2 REST API) for Retrieval-Augmented Generation. + +## Installation + +```xml + + com.google.genkit + genkit-plugin-chroma + 1.0.0-SNAPSHOT + +``` + +## Requirements + +- A running Chroma server (`docker run -p 8000:8000 chromadb/chroma`) +- Java 21+ +- An embedder (e.g. from the Google GenAI plugin) + +## Usage + +`ChromaPlugin` registers a retriever and indexer named `chroma/` for each configured collection. + +```java +import com.google.genkit.plugins.chroma.ChromaCollectionConfig; +import com.google.genkit.plugins.chroma.ChromaPlugin; + +Genkit genkit = Genkit.builder() + .plugin(GoogleGenAIPlugin.create(apiKey)) + .plugin( + ChromaPlugin.builder() + .url("http://localhost:8000") // default + .addCollection( + ChromaCollectionConfig.builder() + .collectionName("films") + .embedderName("googleai/gemini-embedding-001") + .distance(ChromaCollectionConfig.Distance.COSINE) + .createCollectionIfNotExists(true) // default + .build()) + .build()) + .build(); + +// Index and retrieve +genkit.index("chroma/films", documents); +List results = genkit.retrieve("chroma/films", "a Christopher Nolan sci-fi film"); +``` + +## Configuration + +Tune per-collection settings with `ChromaCollectionConfig`: + +- `collectionName` — the Chroma collection name (required) +- `embedderName` — the embedder used to vectorize documents and queries (required) +- `distance` — `COSINE` (default), `L2`, or `INNER_PRODUCT` +- `createCollectionIfNotExists` — auto-create the collection (default `true`) +- `addAdditionalMetadata(key, value)` — metadata merged into every indexed document + +Plugin-level `tenant` and `database` default to `default_tenant` / `default_database`. + +## Sample + +See the [chroma sample](https://github.com/genkit-ai/genkit-java/tree/main/samples/chroma). diff --git a/docs/src/content/docs/plugins/firebase-functions.md b/docs/src/content/docs/plugins/firebase-functions.md index 3057ec16d..fb63e2cfc 100644 --- a/docs/src/content/docs/plugins/firebase-functions.md +++ b/docs/src/content/docs/plugins/firebase-functions.md @@ -49,7 +49,7 @@ public class GeneratePoemFunction implements HttpFunction { genkit.defineFlow("generatePoem", String.class, String.class, (ctx, topic) -> genkit.generate(GenerateOptions.builder() - .model("googleai/gemini-2.0-flash") + .model("googleai/gemini-2.5-flash") .prompt("Write a poem about: " + topic) .build()).getText()); diff --git a/docs/src/content/docs/plugins/firebase-vector-store.md b/docs/src/content/docs/plugins/firebase-vector-store.md index 9d15a1548..a65ab8325 100644 --- a/docs/src/content/docs/plugins/firebase-vector-store.md +++ b/docs/src/content/docs/plugins/firebase-vector-store.md @@ -28,7 +28,7 @@ Genkit genkit = Genkit.builder() .addRetriever(FirestoreRetrieverConfig.builder() .name("my-docs") .collection("documents") - .embedderName("googleai/text-embedding-004") + .embedderName("googleai/gemini-embedding-001") .vectorField("embedding") .contentField("content") .distanceMeasure(FirestoreRetrieverConfig.DistanceMeasure.COSINE) @@ -62,7 +62,7 @@ List results = genkit.retrieve("firebase/my-docs", "What is Genkit?"); genkit.defineFlow("ragQuery", String.class, String.class, (ctx, question) -> { List context = genkit.retrieve("firebase/my-docs", question); return genkit.generate(GenerateOptions.builder() - .model("googleai/gemini-2.0-flash") + .model("googleai/gemini-2.5-flash") .prompt(question) .docs(context) .build()).getText(); @@ -77,7 +77,7 @@ The plugin can automatically create the Firestore database and the required comp FirestoreRetrieverConfig.builder() .name("my-docs") .collection("documents") - .embedderName("googleai/text-embedding-004") + .embedderName("googleai/gemini-embedding-001") .vectorField("embedding") .contentField("content") .createDatabaseIfNotExists(true) @@ -100,7 +100,7 @@ gcloud firestore indexes composite create \ |--------|---------|-------------| | `name` | (required) | Retriever name — used as `firebase/{name}` | | `collection` | (required) | Firestore collection name | -| `embedderName` | (required) | Embedder to use (e.g., `googleai/text-embedding-004`) | +| `embedderName` | (required) | Embedder to use (e.g., `googleai/gemini-embedding-001`) | | `vectorField` | `"embedding"` | Firestore field storing the vector | | `contentField` | `"content"` | Firestore field storing the text content | | `distanceMeasure` | `COSINE` | `COSINE`, `EUCLIDEAN`, or `DOT_PRODUCT` | diff --git a/docs/src/content/docs/plugins/google-genai.md b/docs/src/content/docs/plugins/google-genai.md index 1f32e9a15..f3a5224f3 100644 --- a/docs/src/content/docs/plugins/google-genai.md +++ b/docs/src/content/docs/plugins/google-genai.md @@ -44,7 +44,7 @@ Genkit genkit = Genkit.builder() ModelResponse response = genkit.generate( GenerateOptions.builder() - .model("googleai/gemini-2.0-flash") + .model("googleai/gemini-2.5-flash") .prompt("Tell me about AI") .build()); ``` @@ -62,7 +62,7 @@ List documents = List.of( Document.fromText("Firebase provides cloud services") ); -EmbedResponse response = genkit.embed("googleai/text-embedding-004", documents); +EmbedResponse response = genkit.embed("googleai/gemini-embedding-001", documents); // Access embedding vectors float[] vector = response.getEmbeddings().get(0).getValues(); diff --git a/docs/src/content/docs/plugins/mongodb.md b/docs/src/content/docs/plugins/mongodb.md new file mode 100644 index 000000000..cb4cbfdde --- /dev/null +++ b/docs/src/content/docs/plugins/mongodb.md @@ -0,0 +1,83 @@ +--- +title: MongoDB +description: MongoDB integration for Genkit - Atlas Vector Search retrieval and agent session persistence. +--- + +The MongoDB plugin provides two capabilities: `MongoPlugin` registers Atlas Vector Search retrievers/indexers for RAG, and `MongoSessionStore` persists server-managed agent sessions. + +## Installation + +```xml + + com.google.genkit + genkit-plugin-mongodb + 1.0.0-SNAPSHOT + +``` + +## Requirements + +- MongoDB 4.4+ (Atlas Vector Search requires MongoDB Atlas or the `mongodb/mongodb-atlas-local` Docker image) +- Java 21+ + +## Vector search + +`MongoPlugin` registers a retriever and indexer named `mongodb/` for each configured collection, backed by an [Atlas Vector Search](https://www.mongodb.com/docs/atlas/atlas-vector-search/) index and the `$vectorSearch` aggregation stage. A plain MongoDB server does **not** support `$vectorSearch` — use a MongoDB Atlas cluster or the `mongodb/mongodb-atlas-local` Docker image. + +```java +import com.google.genkit.plugins.mongodb.MongoPlugin; +import com.google.genkit.plugins.mongodb.MongoVectorStoreConfig; + +Genkit genkit = Genkit.builder() + .plugin(GoogleGenAIPlugin.create(apiKey)) + .plugin( + MongoPlugin.builder() + .connectionString("mongodb://localhost:27017/?directConnection=true") + .addCollection( + MongoVectorStoreConfig.builder() + .collectionName("films") + .embedderName("googleai/gemini-embedding-001") + .dimension(768) + .similarity(MongoVectorStoreConfig.Similarity.COSINE) + .createIndexIfNotExists(true) // create the Atlas Vector Search index on first use + .build()) + .build()) + .build(); + +// Index and retrieve +genkit.index("mongodb/films", documents); +List results = genkit.retrieve("mongodb/films", "a Christopher Nolan sci-fi film"); +``` + +Tune per-collection settings with `MongoVectorStoreConfig` (database/collection names, embedder, index name, dimension, similarity — `COSINE`/`EUCLIDEAN`/`DOT_PRODUCT`, text/embedding field names, `numCandidates`, `createIndexIfNotExists`). See the [mongo-vector sample](https://github.com/genkit-ai/genkit-java/tree/main/samples/mongo-vector). + +## Session store + +Construct a `MongoSessionStore` from a `com.mongodb.client.MongoClient` and pass it to an agent's `.store(...)` to persist server-managed sessions in MongoDB: + +```java +import com.google.genkit.plugins.mongodb.session.MongoSessionStore; +import com.google.genkit.plugins.mongodb.session.MongoSessionStoreOptions; +import com.mongodb.client.MongoClient; +import com.mongodb.client.MongoClients; + +MongoClient client = MongoClients.create("mongodb://localhost:27017"); + +MongoSessionStore> store = + new MongoSessionStore<>(client); // uses database "genkit", collection "genkit_sessions" + +Agent> agent = genkit.beta().defineAgent( + AgentConfig.>builder() + .name("myAgent") + .system("You are a helpful assistant.") + .store(store) + .build()); +``` + +All records live in a single collection whose `_id` is `::`; the database and collection are created automatically on first write. It uses the same sharded checkpoint + RFC-6902 diff + pointer layout as the Firestore, DynamoDB, Cosmos DB, and PostgreSQL backends, and supports `onSnapshotStateChange` (via polling), so `chat.abort()` works. + +See [Session Stores](../../agents/session-stores#mongosessionstore) for options. + +## Sample + +See the [agents-mongo-session sample](https://github.com/genkit-ai/genkit-java/tree/main/samples/agents-mongo-session) and the [mongo-vector sample](https://github.com/genkit-ai/genkit-java/tree/main/samples/mongo-vector). diff --git a/docs/src/content/docs/plugins/opentelemetry.md b/docs/src/content/docs/plugins/opentelemetry.md index 45a01cfc3..6235a7ebc 100644 --- a/docs/src/content/docs/plugins/opentelemetry.md +++ b/docs/src/content/docs/plugins/opentelemetry.md @@ -124,7 +124,7 @@ public class App { .build(); // All generate/flow/tool calls now export traces and metrics - genkit.generate(options -> options.model("googleai/gemini-2.0-flash").prompt("Hi")); + genkit.generate(options -> options.model("googleai/gemini-2.5-flash").prompt("Hi")); } } ``` diff --git a/docs/src/content/docs/plugins/postgresql.md b/docs/src/content/docs/plugins/postgresql.md index 472d392ba..ed1a2bd9e 100644 --- a/docs/src/content/docs/plugins/postgresql.md +++ b/docs/src/content/docs/plugins/postgresql.md @@ -38,6 +38,21 @@ Genkit genkit = Genkit.builder() - Batch indexing - Metadata support +## Session store + +The plugin also ships `PostgresSessionStore`, a PostgreSQL-backed agent session store. Construct it from a `javax.sql.DataSource` and pass it to an agent's `.store(...)` to persist server-managed sessions in PostgreSQL: + +```java +import com.google.genkit.plugins.postgresql.session.PostgresSessionStore; +import com.google.genkit.plugins.postgresql.session.PostgresSessionStoreOptions; + +PostgresSessionStore> store = + new PostgresSessionStore<>( + dataSource, PostgresSessionStoreOptions.builder().createTableIfNotExists(true).build()); +``` + +See [Session Stores](../../agents/session-stores#postgressessionstore) for options and the agents-postgres-session sample. + ## Sample See the [postgresql sample](https://github.com/genkit-ai/genkit-java/tree/main/samples/postgresql). diff --git a/docs/src/content/docs/plugins/qdrant.md b/docs/src/content/docs/plugins/qdrant.md new file mode 100644 index 000000000..905976882 --- /dev/null +++ b/docs/src/content/docs/plugins/qdrant.md @@ -0,0 +1,70 @@ +--- +title: Qdrant +description: Qdrant vector database integration for Genkit RAG workflows. +--- + +The Qdrant plugin registers retrievers and indexers backed by a [Qdrant](https://qdrant.tech/) server (REST API) for Retrieval-Augmented Generation. + +## Installation + +```xml + + com.google.genkit + genkit-plugin-qdrant + 1.0.0-SNAPSHOT + +``` + +## Requirements + +- A running Qdrant server (`docker run -p 6333:6333 qdrant/qdrant`) +- Java 21+ +- An embedder (e.g. from the Google GenAI plugin) + +## Usage + +`QdrantPlugin` registers a retriever and indexer named `qdrant/` for each configured collection. + +```java +import com.google.genkit.plugins.qdrant.QdrantCollectionConfig; +import com.google.genkit.plugins.qdrant.QdrantPlugin; + +Genkit genkit = Genkit.builder() + .plugin(GoogleGenAIPlugin.create(apiKey)) + .plugin( + QdrantPlugin.builder() + .url("http://localhost:6333") // default + .apiKey(System.getenv("QDRANT_API_KEY")) // optional; required for Qdrant Cloud + .addCollection( + QdrantCollectionConfig.builder() + .collectionName("films") + .embedderName("googleai/gemini-embedding-001") + .dimension(768) + .distance(QdrantCollectionConfig.Distance.COSINE) + .createCollectionIfNotExists(true) // default + .build()) + .build()) + .build(); + +// Index and retrieve +genkit.index("qdrant/films", documents); +List results = genkit.retrieve("qdrant/films", "a Christopher Nolan sci-fi film"); +``` + +## Configuration + +Tune per-collection settings with `QdrantCollectionConfig`: + +- `collectionName` — the Qdrant collection name (required) +- `embedderName` — the embedder used to vectorize documents and queries (required) +- `dimension` — the embedding dimension used when creating the collection (default `768`). The store also probes the embedder on first use and creates the collection with the model's actual output dimension, so this is only a fallback. +- `distance` — `COSINE` (default), `EUCLIDEAN`, or `DOT_PRODUCT` +- `textPayloadKey` — the payload key that stores the document text (default `text`) +- `createCollectionIfNotExists` — auto-create the collection (default `true`) +- `addAdditionalMetadata(key, value)` — metadata merged into every indexed point's payload + +Document ids that are not integers or UUIDs are hashed into a stable UUID (Qdrant point ids must be an unsigned integer or a UUID). + +## Sample + +See the [qdrant sample](https://github.com/genkit-ai/genkit-java/tree/main/samples/qdrant). diff --git a/genkit/src/main/java/com/google/genkit/Genkit.java b/genkit/src/main/java/com/google/genkit/Genkit.java index 5d010a320..743d93fd4 100644 --- a/genkit/src/main/java/com/google/genkit/Genkit.java +++ b/genkit/src/main/java/com/google/genkit/Genkit.java @@ -2099,7 +2099,7 @@ public GenkitOptions getOptions() { * AgentConfig.builder() * .name("helper") * .system("You are helpful.") - * .model("googleai/gemini-2.0-flash") + * .model("googleai/gemini-2.5-flash") * .build()); * } * diff --git a/plugins/chroma/pom.xml b/plugins/chroma/pom.xml new file mode 100644 index 000000000..879aac02d --- /dev/null +++ b/plugins/chroma/pom.xml @@ -0,0 +1,78 @@ + + + + 4.0.0 + + + com.google.genkit + genkit-parent + 1.0.0-SNAPSHOT + ../../pom.xml + + + genkit-plugin-chroma + jar + Genkit Chroma Plugin + Chroma vector database integration for Genkit - indexer and retriever for RAG workflows + + + false + + + + + + com.google.genkit + genkit-core + ${project.version} + + + com.google.genkit + genkit-ai + ${project.version} + + + + + com.fasterxml.jackson.core + jackson-databind + + + + + org.slf4j + slf4j-api + + + + + org.junit.jupiter + junit-jupiter + test + + + org.mockito + mockito-core + test + + + diff --git a/plugins/chroma/src/main/java/com/google/genkit/plugins/chroma/ChromaCollectionConfig.java b/plugins/chroma/src/main/java/com/google/genkit/plugins/chroma/ChromaCollectionConfig.java new file mode 100644 index 000000000..1e04d499a --- /dev/null +++ b/plugins/chroma/src/main/java/com/google/genkit/plugins/chroma/ChromaCollectionConfig.java @@ -0,0 +1,206 @@ +/* + * Copyright 2025 Google LLC + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * + * SPDX-License-Identifier: Apache-2.0 + */ + +package com.google.genkit.plugins.chroma; + +import java.util.HashMap; +import java.util.Map; + +/** + * Configuration for a single Chroma collection managed by {@link ChromaPlugin}. + * + *

Each config registers a retriever and indexer named {@code chroma/}. + */ +public final class ChromaCollectionConfig { + + /** Distance function used by the Chroma HNSW index. */ + public enum Distance { + COSINE("cosine"), + L2("l2"), + INNER_PRODUCT("ip"); + + private final String value; + + Distance(String value) { + this.value = value; + } + + /** + * Returns the Chroma {@code hnsw:space} value. + * + * @return the space name + */ + public String getValue() { + return value; + } + } + + private final String collectionName; + private final String embedderName; + private final Distance distance; + private final boolean createCollectionIfNotExists; + private final Map additionalMetadata; + + private ChromaCollectionConfig(Builder builder) { + this.collectionName = builder.collectionName; + this.embedderName = builder.embedderName; + this.distance = builder.distance; + this.createCollectionIfNotExists = builder.createCollectionIfNotExists; + this.additionalMetadata = new HashMap<>(builder.additionalMetadata); + } + + /** + * Returns the Chroma collection name. + * + * @return the collection name + */ + public String getCollectionName() { + return collectionName; + } + + /** + * Returns the name of the embedder used to vectorize documents and queries. + * + * @return the embedder name + */ + public String getEmbedderName() { + return embedderName; + } + + /** + * Returns the distance function (default {@link Distance#COSINE}). + * + * @return the distance function + */ + public Distance getDistance() { + return distance; + } + + /** + * Returns whether to create the collection on first use if it does not exist (default {@code + * true}). + * + * @return {@code true} if the collection should be created when missing + */ + public boolean isCreateCollectionIfNotExists() { + return createCollectionIfNotExists; + } + + /** + * Returns additional metadata merged into every indexed document. + * + * @return the additional metadata + */ + public Map getAdditionalMetadata() { + return additionalMetadata; + } + + /** + * Creates a new builder. + * + * @return a new builder + */ + public static Builder builder() { + return new Builder(); + } + + /** Builder for {@link ChromaCollectionConfig}. */ + public static final class Builder { + private String collectionName; + private String embedderName; + private Distance distance = Distance.COSINE; + private boolean createCollectionIfNotExists = true; + private final Map additionalMetadata = new HashMap<>(); + + private Builder() {} + + /** + * Sets the collection name. + * + * @param collectionName the collection name + * @return this builder + */ + public Builder collectionName(String collectionName) { + this.collectionName = collectionName; + return this; + } + + /** + * Sets the embedder name. + * + * @param embedderName the embedder name + * @return this builder + */ + public Builder embedderName(String embedderName) { + this.embedderName = embedderName; + return this; + } + + /** + * Sets the distance function. + * + * @param distance the distance function + * @return this builder + */ + public Builder distance(Distance distance) { + this.distance = distance; + return this; + } + + /** + * Sets whether to create the collection on first use if it does not exist. + * + * @param createCollectionIfNotExists whether to create the collection when missing + * @return this builder + */ + public Builder createCollectionIfNotExists(boolean createCollectionIfNotExists) { + this.createCollectionIfNotExists = createCollectionIfNotExists; + return this; + } + + /** + * Adds a metadata entry merged into every indexed document. + * + * @param key the metadata key + * @param value the metadata value + * @return this builder + */ + public Builder addAdditionalMetadata(String key, Object value) { + this.additionalMetadata.put(key, value); + return this; + } + + /** + * Builds a new {@code ChromaCollectionConfig}. + * + * @return a new config instance + */ + public ChromaCollectionConfig build() { + if (collectionName == null || collectionName.isBlank()) { + throw new IllegalArgumentException("collectionName must be non-empty"); + } + if (embedderName == null || embedderName.isBlank()) { + throw new IllegalArgumentException("embedderName must be non-empty"); + } + if (distance == null) { + throw new IllegalArgumentException("distance must be non-null"); + } + return new ChromaCollectionConfig(this); + } + } +} diff --git a/plugins/chroma/src/main/java/com/google/genkit/plugins/chroma/ChromaPlugin.java b/plugins/chroma/src/main/java/com/google/genkit/plugins/chroma/ChromaPlugin.java new file mode 100644 index 000000000..06cd45fd1 --- /dev/null +++ b/plugins/chroma/src/main/java/com/google/genkit/plugins/chroma/ChromaPlugin.java @@ -0,0 +1,201 @@ +/* + * Copyright 2025 Google LLC + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * + * SPDX-License-Identifier: Apache-2.0 + */ + +package com.google.genkit.plugins.chroma; + +import com.google.genkit.ai.Embedder; +import com.google.genkit.core.Action; +import com.google.genkit.core.ActionType; +import com.google.genkit.core.Plugin; +import com.google.genkit.core.Registry; +import java.util.ArrayList; +import java.util.List; +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + +/** + * Chroma vector database plugin for Genkit. + * + *

Registers a retriever and indexer named {@code chroma/} for each configured + * collection, talking to a Chroma server over its v2 REST API. + * + *

Example usage: + * + * 

{@code
+ * Genkit genkit = Genkit.builder()
+ *     .plugin(GoogleGenAIPlugin.create(apiKey))
+ *     .plugin(
+ *         ChromaPlugin.builder()
+ *             .url("http://localhost:8000")
+ *             .addCollection(
+ *                 ChromaCollectionConfig.builder()
+ *                     .collectionName("films")
+ *                     .embedderName("googleai/gemini-embedding-001")
+ *                     .build())
+ *             .build())
+ *     .build();
+ * }
+ */ +public final class ChromaPlugin implements Plugin { + + /** The plugin name; used as the {@code chroma/...} action prefix. */ + public static final String PLUGIN_NAME = "chroma"; + + /** Default Chroma tenant. */ + public static final String DEFAULT_TENANT = "default_tenant"; + + /** Default Chroma database. */ + public static final String DEFAULT_DATABASE = "default_database"; + + /** Default Chroma server URL. */ + public static final String DEFAULT_URL = "http://localhost:8000"; + + private static final Logger logger = LoggerFactory.getLogger(ChromaPlugin.class); + + private final String url; + private final String tenant; + private final String database; + private final List collectionConfigs; + + private ChromaPlugin(Builder builder) { + this.url = builder.url; + this.tenant = builder.tenant; + this.database = builder.database; + this.collectionConfigs = new ArrayList<>(builder.collectionConfigs); + } + + /** + * Creates a new builder. + * + * @return a new builder + */ + public static Builder builder() { + return new Builder(); + } + + @Override + public String getName() { + return PLUGIN_NAME; + } + + @Override + public List> init() { + throw new IllegalStateException( + "ChromaPlugin requires a Registry to resolve embedders. Use init(registry) instead."); + } + + @Override + public List> init(Registry registry) { + List> actions = new ArrayList<>(); + for (ChromaCollectionConfig config : collectionConfigs) { + String embedderKey = ActionType.EMBEDDER.keyFromName(config.getEmbedderName()); + Action embedderAction = registry.lookupAction(embedderKey); + if (embedderAction == null) { + throw new IllegalStateException( + "Embedder not found: " + + config.getEmbedderName() + + ". Make sure the embedder plugin is registered before ChromaPlugin."); + } + if (!(embedderAction instanceof Embedder embedder)) { + throw new IllegalStateException( + "Action " + config.getEmbedderName() + " is not an Embedder"); + } + + ChromaVectorStore store = new ChromaVectorStore(url, tenant, database, config, embedder); + actions.add(store.createRetriever()); + actions.add(store.createIndexer()); + logger.info("Registered Chroma vector store: {}/{}", PLUGIN_NAME, config.getCollectionName()); + } + return actions; + } + + /** Builder for {@link ChromaPlugin}. */ + public static final class Builder { + private String url = DEFAULT_URL; + private String tenant = DEFAULT_TENANT; + private String database = DEFAULT_DATABASE; + private final List collectionConfigs = new ArrayList<>(); + + private Builder() {} + + /** + * Sets the Chroma server URL (default {@value #DEFAULT_URL}). + * + * @param url the server URL + * @return this builder + */ + public Builder url(String url) { + this.url = url; + return this; + } + + /** + * Sets the Chroma tenant (default {@value #DEFAULT_TENANT}). + * + * @param tenant the tenant + * @return this builder + */ + public Builder tenant(String tenant) { + this.tenant = tenant; + return this; + } + + /** + * Sets the Chroma database (default {@value #DEFAULT_DATABASE}). + * + * @param database the database + * @return this builder + */ + public Builder database(String database) { + this.database = database; + return this; + } + + /** + * Adds a collection configuration. + * + * @param config the collection configuration + * @return this builder + */ + public Builder addCollection(ChromaCollectionConfig config) { + this.collectionConfigs.add(config); + return this; + } + + /** + * Builds the plugin. + * + * @return a new {@code ChromaPlugin} + */ + public ChromaPlugin build() { + if (url == null || url.isBlank()) { + throw new IllegalStateException("url is required"); + } + if (tenant == null || tenant.isBlank()) { + throw new IllegalStateException("tenant is required"); + } + if (database == null || database.isBlank()) { + throw new IllegalStateException("database is required"); + } + if (collectionConfigs.isEmpty()) { + throw new IllegalStateException("At least one collection configuration is required"); + } + return new ChromaPlugin(this); + } + } +} diff --git a/plugins/chroma/src/main/java/com/google/genkit/plugins/chroma/ChromaVectorStore.java b/plugins/chroma/src/main/java/com/google/genkit/plugins/chroma/ChromaVectorStore.java new file mode 100644 index 000000000..ea1700cdd --- /dev/null +++ b/plugins/chroma/src/main/java/com/google/genkit/plugins/chroma/ChromaVectorStore.java @@ -0,0 +1,311 @@ +/* + * Copyright 2025 Google LLC + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * + * SPDX-License-Identifier: Apache-2.0 + */ + +package com.google.genkit.plugins.chroma; + +import com.fasterxml.jackson.databind.JsonNode; +import com.fasterxml.jackson.databind.ObjectMapper; +import com.fasterxml.jackson.databind.node.ArrayNode; +import com.fasterxml.jackson.databind.node.ObjectNode; +import com.google.genkit.ai.Document; +import com.google.genkit.ai.EmbedRequest; +import com.google.genkit.ai.EmbedResponse; +import com.google.genkit.ai.Embedder; +import com.google.genkit.ai.Indexer; +import com.google.genkit.ai.IndexerRequest; +import com.google.genkit.ai.IndexerResponse; +import com.google.genkit.ai.Retriever; +import com.google.genkit.ai.RetrieverRequest; +import com.google.genkit.ai.RetrieverResponse; +import com.google.genkit.core.ActionContext; +import java.net.URI; +import java.net.http.HttpClient; +import java.net.http.HttpRequest; +import java.net.http.HttpResponse; +import java.nio.charset.StandardCharsets; +import java.util.ArrayList; +import java.util.HashMap; +import java.util.List; +import java.util.Map; +import java.util.UUID; +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + +/** + * Chroma vector store backed by the Chroma v2 REST API. + * + *

Indexes documents (id + text + embedding + metadata) into a Chroma collection and retrieves + * the nearest neighbors of a query embedding. + */ +public final class ChromaVectorStore { + + private static final Logger logger = LoggerFactory.getLogger(ChromaVectorStore.class); + private static final ObjectMapper MAPPER = new ObjectMapper(); + + private final HttpClient http = HttpClient.newHttpClient(); + private final String baseUrl; + private final String tenant; + private final String database; + private final ChromaCollectionConfig config; + private final Embedder embedder; + + private volatile String collectionId; + + /** + * Creates a new store. + * + * @param baseUrl the Chroma server base URL (e.g. {@code http://localhost:8000}) + * @param tenant the Chroma tenant + * @param database the Chroma database + * @param config the collection configuration + * @param embedder the embedder used to vectorize documents and queries + */ + public ChromaVectorStore( + String baseUrl, + String tenant, + String database, + ChromaCollectionConfig config, + Embedder embedder) { + this.baseUrl = baseUrl.endsWith("/") ? baseUrl.substring(0, baseUrl.length() - 1) : baseUrl; + this.tenant = tenant; + this.database = database; + this.config = config; + this.embedder = embedder; + } + + /** Creates the retriever action registered by the plugin. */ + Retriever createRetriever() { + String name = ChromaPlugin.PLUGIN_NAME + "/" + config.getCollectionName(); + return Retriever.builder().name(name).handler(this::retrieve).build(); + } + + /** Creates the indexer action registered by the plugin. */ + Indexer createIndexer() { + String name = ChromaPlugin.PLUGIN_NAME + "/" + config.getCollectionName(); + return Indexer.builder().name(name).handler(this::index).build(); + } + + private String collectionsPath() { + return "/api/v2/tenants/" + tenant + "/databases/" + database + "/collections"; + } + + private synchronized String ensureCollection() { + if (collectionId != null) { + return collectionId; + } + ObjectNode body = MAPPER.createObjectNode(); + body.put("name", config.getCollectionName()); + body.put("get_or_create", config.isCreateCollectionIfNotExists()); + ObjectNode metadata = body.putObject("metadata"); + metadata.put("hnsw:space", config.getDistance().getValue()); + + JsonNode resp = send("POST", collectionsPath(), body); + JsonNode id = resp.get("id"); + if (id == null || id.isNull()) { + throw new RuntimeException( + "Chroma did not return a collection id for " + config.getCollectionName()); + } + collectionId = id.asText(); + logger.info("Using Chroma collection {} ({})", config.getCollectionName(), collectionId); + return collectionId; + } + + /** + * Retrieves documents similar to the query. + * + * @param context the action context + * @param request the retriever request + * @return the retriever response with matching documents + */ + public RetrieverResponse retrieve(ActionContext context, RetrieverRequest request) { + Document queryDoc = request.getQuery(); + if (queryDoc == null || queryDoc.text() == null || queryDoc.text().isBlank()) { + throw new RuntimeException("Query document has no text content"); + } + int topK = + request.getOptions() != null && request.getOptions().getK() != null + ? request.getOptions().getK() + : 10; + List queryEmbedding = generateEmbedding(context, queryDoc.text()); + String id = ensureCollection(); + + ObjectNode body = MAPPER.createObjectNode(); + ArrayNode queryEmbeddings = body.putArray("query_embeddings"); + queryEmbeddings.add(floatsToArray(queryEmbedding)); + body.put("n_results", topK); + ArrayNode include = body.putArray("include"); + include.add("documents"); + include.add("metadatas"); + include.add("distances"); + + JsonNode resp = send("POST", collectionsPath() + "/" + id + "/query", body); + List documents = new ArrayList<>(); + JsonNode idsOuter = resp.get("ids"); + if (idsOuter == null || !idsOuter.isArray() || idsOuter.isEmpty()) { + return new RetrieverResponse(documents); + } + JsonNode ids = idsOuter.get(0); + JsonNode docs = firstRow(resp.get("documents")); + JsonNode metadatas = firstRow(resp.get("metadatas")); + JsonNode distances = firstRow(resp.get("distances")); + for (int i = 0; i < ids.size(); i++) { + String content = + docs != null && docs.get(i) != null && !docs.get(i).isNull() ? docs.get(i).asText() : ""; + Map metadata = new HashMap<>(); + if (metadatas != null && metadatas.get(i) != null && metadatas.get(i).isObject()) { + metadata.putAll(MAPPER.convertValue(metadatas.get(i), Map.class)); + } + metadata.put("id", ids.get(i).asText()); + if (distances != null && distances.get(i) != null && distances.get(i).isNumber()) { + double distance = distances.get(i).asDouble(); + metadata.put("distance", distance); + metadata.put("score", 1.0 - distance); + } + Document doc = new Document(content); + doc.setMetadata(metadata); + documents.add(doc); + } + logger.debug( + "Retrieved {} documents from collection {}", documents.size(), config.getCollectionName()); + return new RetrieverResponse(documents); + } + + /** + * Indexes documents into the collection, generating an embedding for each. + * + * @param context the action context + * @param request the indexer request + * @return the indexer response + */ + public IndexerResponse index(ActionContext context, IndexerRequest request) { + List documents = request.getDocuments(); + if (documents == null || documents.isEmpty()) { + logger.warn("No documents to index"); + return new IndexerResponse(); + } + String id = ensureCollection(); + + ObjectNode body = MAPPER.createObjectNode(); + ArrayNode ids = body.putArray("ids"); + ArrayNode embeddings = body.putArray("embeddings"); + ArrayNode contents = body.putArray("documents"); + ArrayNode metadatas = body.putArray("metadatas"); + + for (Document doc : documents) { + String content = doc.text() != null ? doc.text() : ""; + List embedding = generateEmbedding(context, content); + ids.add(getOrGenerateId(doc)); + embeddings.add(floatsToArray(embedding)); + contents.add(content); + + Map metadata = new HashMap<>(); + if (doc.getMetadata() != null) { + for (Map.Entry entry : doc.getMetadata().entrySet()) { + if (!"id".equals(entry.getKey())) { + metadata.put(entry.getKey(), entry.getValue()); + } + } + } + metadata.putAll(config.getAdditionalMetadata()); + // Chroma rejects empty metadata objects; send null when there is nothing to store. + if (metadata.isEmpty()) { + metadatas.addNull(); + } else { + metadatas.add(MAPPER.valueToTree(metadata)); + } + } + + send("POST", collectionsPath() + "/" + id + "/add", body); + logger.info( + "Indexed {} documents into collection {}", documents.size(), config.getCollectionName()); + return new IndexerResponse(); + } + + private static JsonNode firstRow(JsonNode outer) { + return (outer != null && outer.isArray() && !outer.isEmpty()) ? outer.get(0) : null; + } + + private ArrayNode floatsToArray(List values) { + ArrayNode arr = MAPPER.createArrayNode(); + for (float v : values) { + arr.add(v); + } + return arr; + } + + private List generateEmbedding(ActionContext ctx, String text) { + EmbedResponse response = embedder.run(ctx, new EmbedRequest(List.of(new Document(text)))); + if (response.getEmbeddings() == null || response.getEmbeddings().isEmpty()) { + throw new RuntimeException("Failed to generate embedding for text"); + } + float[] values = response.getEmbeddings().get(0).getValues(); + List out = new ArrayList<>(values.length); + for (float v : values) { + out.add(v); + } + return out; + } + + private String getOrGenerateId(Document doc) { + if (doc.getMetadata() != null && doc.getMetadata().get("id") != null) { + return doc.getMetadata().get("id").toString(); + } + return UUID.randomUUID().toString(); + } + + private JsonNode send(String method, String path, JsonNode body) { + try { + HttpRequest.Builder builder = + HttpRequest.newBuilder() + .uri(URI.create(baseUrl + path)) + .header("Content-Type", "application/json") + .header("Accept", "application/json"); + if (body != null) { + builder.method( + method, + HttpRequest.BodyPublishers.ofString( + MAPPER.writeValueAsString(body), StandardCharsets.UTF_8)); + } else { + builder.method(method, HttpRequest.BodyPublishers.noBody()); + } + HttpResponse response = + http.send(builder.build(), HttpResponse.BodyHandlers.ofString()); + if (response.statusCode() / 100 != 2) { + throw new RuntimeException( + "Chroma request " + + method + + " " + + path + + " failed (" + + response.statusCode() + + "): " + + response.body()); + } + String responseBody = response.body(); + if (responseBody == null || responseBody.isBlank()) { + return MAPPER.createObjectNode(); + } + return MAPPER.readTree(responseBody); + } catch (RuntimeException e) { + throw e; + } catch (Exception e) { + throw new RuntimeException( + "Chroma request " + method + " " + path + " failed: " + e.getMessage(), e); + } + } +} diff --git a/plugins/chroma/src/main/java/com/google/genkit/plugins/chroma/package-info.java b/plugins/chroma/src/main/java/com/google/genkit/plugins/chroma/package-info.java new file mode 100644 index 000000000..2cb3b0058 --- /dev/null +++ b/plugins/chroma/src/main/java/com/google/genkit/plugins/chroma/package-info.java @@ -0,0 +1,25 @@ +/* + * Copyright 2025 Google LLC + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * + * SPDX-License-Identifier: Apache-2.0 + */ + +/** + * Chroma vector database integration for Genkit. + * + *

{@link com.google.genkit.plugins.chroma.ChromaPlugin} registers retrievers and indexers backed + * by a Chroma server (v2 REST API) for RAG workflows. + */ +package com.google.genkit.plugins.chroma; diff --git a/plugins/chroma/src/test/java/com/google/genkit/plugins/chroma/ChromaCollectionConfigTest.java b/plugins/chroma/src/test/java/com/google/genkit/plugins/chroma/ChromaCollectionConfigTest.java new file mode 100644 index 000000000..07a425916 --- /dev/null +++ b/plugins/chroma/src/test/java/com/google/genkit/plugins/chroma/ChromaCollectionConfigTest.java @@ -0,0 +1,70 @@ +/* + * Copyright 2025 Google LLC + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * + * SPDX-License-Identifier: Apache-2.0 + */ + +package com.google.genkit.plugins.chroma; + +import static org.junit.jupiter.api.Assertions.assertEquals; +import static org.junit.jupiter.api.Assertions.assertThrows; +import static org.junit.jupiter.api.Assertions.assertTrue; + +import org.junit.jupiter.api.Test; + +/** Unit tests for {@link ChromaCollectionConfig}. */ +class ChromaCollectionConfigTest { + + @Test + void defaultsAreSane() { + ChromaCollectionConfig c = + ChromaCollectionConfig.builder() + .collectionName("films") + .embedderName("googleai/gemini-embedding-001") + .build(); + assertEquals("films", c.getCollectionName()); + assertEquals("googleai/gemini-embedding-001", c.getEmbedderName()); + assertEquals(ChromaCollectionConfig.Distance.COSINE, c.getDistance()); + assertEquals("cosine", c.getDistance().getValue()); + assertTrue(c.isCreateCollectionIfNotExists()); + assertTrue(c.getAdditionalMetadata().isEmpty()); + } + + @Test + void customBuilder() { + ChromaCollectionConfig c = + ChromaCollectionConfig.builder() + .collectionName("docs") + .embedderName("e") + .distance(ChromaCollectionConfig.Distance.L2) + .createCollectionIfNotExists(false) + .addAdditionalMetadata("source", "wiki") + .build(); + assertEquals(ChromaCollectionConfig.Distance.L2, c.getDistance()); + assertEquals("l2", c.getDistance().getValue()); + assertEquals(false, c.isCreateCollectionIfNotExists()); + assertEquals("wiki", c.getAdditionalMetadata().get("source")); + } + + @Test + void builderValidates() { + assertThrows( + IllegalArgumentException.class, + () -> ChromaCollectionConfig.builder().embedderName("e").build()); + assertThrows( + IllegalArgumentException.class, + () -> ChromaCollectionConfig.builder().collectionName("c").build()); + } +} diff --git a/plugins/chroma/src/test/java/com/google/genkit/plugins/chroma/ChromaPluginTest.java b/plugins/chroma/src/test/java/com/google/genkit/plugins/chroma/ChromaPluginTest.java new file mode 100644 index 000000000..5eff068e3 --- /dev/null +++ b/plugins/chroma/src/test/java/com/google/genkit/plugins/chroma/ChromaPluginTest.java @@ -0,0 +1,55 @@ +/* + * Copyright 2025 Google LLC + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * + * SPDX-License-Identifier: Apache-2.0 + */ + +package com.google.genkit.plugins.chroma; + +import static org.junit.jupiter.api.Assertions.assertEquals; +import static org.junit.jupiter.api.Assertions.assertThrows; + +import org.junit.jupiter.api.Test; + +/** Unit tests for {@link ChromaPlugin}. */ +class ChromaPluginTest { + + private static ChromaCollectionConfig config() { + return ChromaCollectionConfig.builder().collectionName("films").embedderName("e").build(); + } + + @Test + void getName() { + assertEquals("chroma", ChromaPlugin.builder().addCollection(config()).build().getName()); + } + + @Test + void builderDefaults() { + // A valid plugin builds with default url/tenant/database. + ChromaPlugin plugin = ChromaPlugin.builder().addCollection(config()).build(); + assertEquals("chroma", plugin.getName()); + } + + @Test + void requiresAtLeastOneCollection() { + assertThrows(IllegalStateException.class, () -> ChromaPlugin.builder().build()); + } + + @Test + void initWithoutRegistryThrows() { + ChromaPlugin plugin = ChromaPlugin.builder().addCollection(config()).build(); + assertThrows(IllegalStateException.class, plugin::init); + } +} diff --git a/plugins/chroma/src/test/java/com/google/genkit/plugins/chroma/ChromaVectorStoreTest.java b/plugins/chroma/src/test/java/com/google/genkit/plugins/chroma/ChromaVectorStoreTest.java new file mode 100644 index 000000000..5cbd5544b --- /dev/null +++ b/plugins/chroma/src/test/java/com/google/genkit/plugins/chroma/ChromaVectorStoreTest.java @@ -0,0 +1,111 @@ +/* + * Copyright 2025 Google LLC + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * + * SPDX-License-Identifier: Apache-2.0 + */ + +package com.google.genkit.plugins.chroma; + +import static org.junit.jupiter.api.Assertions.assertEquals; +import static org.junit.jupiter.api.Assertions.assertFalse; +import static org.junit.jupiter.api.Assumptions.assumeTrue; + +import com.google.genkit.ai.Document; +import com.google.genkit.ai.EmbedResponse; +import com.google.genkit.ai.Embedder; +import com.google.genkit.ai.EmbedderInfo; +import com.google.genkit.ai.IndexerRequest; +import com.google.genkit.ai.RetrieverRequest; +import com.google.genkit.ai.RetrieverResponse; +import java.util.List; +import java.util.Random; +import java.util.UUID; +import org.junit.jupiter.api.BeforeEach; +import org.junit.jupiter.api.Test; + +/** + * Integration tests for {@link ChromaVectorStore}, gated on the {@code CHROMA_URL} environment + * variable (e.g. a Chroma server started with Docker). Skipped via {@link + * org.junit.jupiter.api.Assumptions} when unset. Uses a deterministic stub embedder so a query + * equal to an indexed document's text retrieves that document first. + */ +class ChromaVectorStoreTest { + + private static final String URL = System.getenv("CHROMA_URL"); + private static final int DIM = 16; + + private ChromaVectorStore store; + + private static boolean configured() { + return URL != null && !URL.isEmpty(); + } + + /** Deterministic embedder: identical text always maps to the same vector. */ + private static Embedder stubEmbedder() { + return new Embedder( + "test/stub", + new EmbedderInfo(), + (ctx, req) -> { + List out = new java.util.ArrayList<>(); + for (Document doc : req.getDocuments()) { + Random random = new Random(doc.text().hashCode()); + float[] values = new float[DIM]; + for (int i = 0; i < DIM; i++) { + values[i] = random.nextFloat(); + } + out.add(new EmbedResponse.Embedding(values)); + } + return new EmbedResponse(out); + }); + } + + @BeforeEach + void setUp() { + if (!configured()) { + return; + } + ChromaCollectionConfig config = + ChromaCollectionConfig.builder() + .collectionName("genkit_test_" + UUID.randomUUID().toString().replace("-", "")) + .embedderName("test/stub") + .build(); + store = + new ChromaVectorStore(URL, "default_tenant", "default_database", config, stubEmbedder()); + } + + @Test + void indexThenRetrieveReturnsNearestFirst() { + assumeTrue(configured()); + List docs = + List.of( + Document.fromText("The Matrix is a sci-fi film about simulated reality."), + Document.fromText("The Godfather is a crime film about a mafia family."), + Document.fromText("Inception is a sci-fi film about dreams.")); + store.index(null, new IndexerRequest(docs)); + + RetrieverRequest request = + new RetrieverRequest( + docs.get(1).text() == null ? null : Document.fromText(docs.get(1).text())); + RetrieverRequest.RetrieverOptions options = new RetrieverRequest.RetrieverOptions(); + options.setK(3); + request.setOptions(options); + + RetrieverResponse response = store.retrieve(null, request); + assertFalse(response.getDocuments().isEmpty()); + assertEquals( + "The Godfather is a crime film about a mafia family.", + response.getDocuments().get(0).text()); + } +} diff --git a/plugins/evaluators/README.md b/plugins/evaluators/README.md index a353211b2..4c7ece9ce 100644 --- a/plugins/evaluators/README.md +++ b/plugins/evaluators/README.md @@ -51,7 +51,7 @@ import com.google.genkit.plugins.evaluators.GenkitMetric; Genkit genkit = Genkit.builder() .addPlugin(EvaluatorsPlugin.create( EvaluatorsPluginOptions.builder() - .judge("googleai/gemini-2.0-flash") + .judge("googleai/gemini-2.5-flash") .metricTypes(List.of( GenkitMetric.FAITHFULNESS, GenkitMetric.ANSWER_RELEVANCY, @@ -67,7 +67,7 @@ Genkit genkit = Genkit.builder() Genkit genkit = Genkit.builder() .addPlugin(EvaluatorsPlugin.create( EvaluatorsPluginOptions.builder() - .judge("googleai/gemini-2.0-flash") + .judge("googleai/gemini-2.5-flash") .useAllMetrics() .build())) .build(); @@ -79,8 +79,8 @@ Genkit genkit = Genkit.builder() Genkit genkit = Genkit.builder() .addPlugin(EvaluatorsPlugin.create( EvaluatorsPluginOptions.builder() - .judge("googleai/gemini-2.0-flash") // Default judge - .embedder("googleai/text-embedding-004") // Default embedder + .judge("googleai/gemini-2.5-flash") // Default judge + .embedder("googleai/gemini-embedding-001") // Default embedder .metrics(List.of( // Use defaults MetricConfig.of(GenkitMetric.FAITHFULNESS), @@ -201,7 +201,7 @@ for (EvalResponse response : results) { Any model registered with Genkit can be used as a judge: -- `googleai/gemini-2.0-flash` +- `googleai/gemini-2.5-flash` - `googleai/gemini-1.5-pro` - `openai/gpt-4o` - `anthropic/claude-3-sonnet` diff --git a/plugins/evaluators/src/main/java/com/google/genkit/plugins/evaluators/EvaluatorsPlugin.java b/plugins/evaluators/src/main/java/com/google/genkit/plugins/evaluators/EvaluatorsPlugin.java index 29f1562f3..ed0ef1a5d 100644 --- a/plugins/evaluators/src/main/java/com/google/genkit/plugins/evaluators/EvaluatorsPlugin.java +++ b/plugins/evaluators/src/main/java/com/google/genkit/plugins/evaluators/EvaluatorsPlugin.java @@ -58,7 +58,7 @@ * .addPlugin( * EvaluatorsPlugin.create( * EvaluatorsPluginOptions.builder() - * .judge("googleai/gemini-2.0-flash") + * .judge("googleai/gemini-2.5-flash") * .metricTypes(List.of(GenkitMetric.FAITHFULNESS, GenkitMetric.ANSWER_RELEVANCY)) * .build())) * .build(); diff --git a/plugins/evaluators/src/test/java/com/google/genkit/plugins/evaluators/EvaluatorsPluginOptionsTest.java b/plugins/evaluators/src/test/java/com/google/genkit/plugins/evaluators/EvaluatorsPluginOptionsTest.java index d35ee90fe..fdccf7ded 100644 --- a/plugins/evaluators/src/test/java/com/google/genkit/plugins/evaluators/EvaluatorsPluginOptionsTest.java +++ b/plugins/evaluators/src/test/java/com/google/genkit/plugins/evaluators/EvaluatorsPluginOptionsTest.java @@ -39,9 +39,9 @@ void testBuilderWithDefaults() { @Test void testBuilderWithJudge() { EvaluatorsPluginOptions options = - EvaluatorsPluginOptions.builder().judge("googleai/gemini-2.0-flash").build(); + EvaluatorsPluginOptions.builder().judge("googleai/gemini-2.5-flash").build(); - assertEquals("googleai/gemini-2.0-flash", options.getJudge()); + assertEquals("googleai/gemini-2.5-flash", options.getJudge()); } @Test diff --git a/plugins/firebase/README.md b/plugins/firebase/README.md index 5bf0fef71..44bd3de9e 100644 --- a/plugins/firebase/README.md +++ b/plugins/firebase/README.md @@ -44,7 +44,7 @@ Genkit genkit = Genkit.builder() .addRetriever(FirestoreRetrieverConfig.builder() .name("myDocs") .collection("documents") - .embedderName("googleai/text-embedding-004") + .embedderName("googleai/gemini-embedding-001") .vectorField("embedding") .contentField("content") .build()) @@ -91,7 +91,7 @@ Genkit genkit = Genkit.builder() .addRetriever(FirestoreRetrieverConfig.builder() .name("myDocs") .collection("documents") - .embedderName("googleai/text-embedding-004") + .embedderName("googleai/gemini-embedding-001") .vectorField("embedding") .contentField("content") .distanceMeasure(FirestoreRetrieverConfig.DistanceMeasure.COSINE) @@ -206,7 +206,7 @@ public class RAGFunction implements HttpFunction { .addRetriever(FirestoreRetrieverConfig.builder() .name("docs") .collection("documents") - .embedderName("googleai/text-embedding-004") + .embedderName("googleai/gemini-embedding-001") .vectorField("embedding") .contentField("content") .build()) @@ -334,7 +334,7 @@ public class FirebaseRAGExample { .addRetriever(FirestoreRetrieverConfig.builder() .name("knowledge-base") .collection("documents") - .embedderName("googleai/text-embedding-004") + .embedderName("googleai/gemini-embedding-001") .vectorField("embedding") .contentField("content") .build()) diff --git a/plugins/firebase/src/main/java/com/google/genkit/plugins/firebase/FirebasePlugin.java b/plugins/firebase/src/main/java/com/google/genkit/plugins/firebase/FirebasePlugin.java index 1fdb17f58..eac41ba83 100644 --- a/plugins/firebase/src/main/java/com/google/genkit/plugins/firebase/FirebasePlugin.java +++ b/plugins/firebase/src/main/java/com/google/genkit/plugins/firebase/FirebasePlugin.java @@ -61,7 +61,7 @@ * FirestoreRetrieverConfig.builder() * .name("my-docs") * .collection("documents") - * .embedderName("googleai/text-embedding-004") + * .embedderName("googleai/gemini-embedding-001") * .vectorField("embedding") * .contentField("content") * .build()) diff --git a/plugins/firebase/src/main/java/com/google/genkit/plugins/firebase/retriever/FirestoreRetrieverConfig.java b/plugins/firebase/src/main/java/com/google/genkit/plugins/firebase/retriever/FirestoreRetrieverConfig.java index ef29ce9a3..dd76b0b74 100644 --- a/plugins/firebase/src/main/java/com/google/genkit/plugins/firebase/retriever/FirestoreRetrieverConfig.java +++ b/plugins/firebase/src/main/java/com/google/genkit/plugins/firebase/retriever/FirestoreRetrieverConfig.java @@ -34,7 +34,7 @@ * FirestoreRetrieverConfig config = FirestoreRetrieverConfig.builder() * .name("my-docs") * .collection("documents") - * .embedderName("googleai/text-embedding-004") + * .embedderName("googleai/gemini-embedding-001") * .vectorField("embedding") * .contentField("content") * .distanceMeasure(DistanceMeasure.COSINE) @@ -328,7 +328,7 @@ public Builder embedder(Embedder embedder) { /** * Sets the embedder name for resolution from registry. * - * @param embedderName the embedder name (e.g., "googleai/text-embedding-004") + * @param embedderName the embedder name (e.g., "googleai/gemini-embedding-001") * @return this builder */ public Builder embedderName(String embedderName) { diff --git a/plugins/google-genai/src/main/java/com/google/genkit/plugins/googlegenai/GeminiModel.java b/plugins/google-genai/src/main/java/com/google/genkit/plugins/googlegenai/GeminiModel.java index 60d818cd0..ad4133fea 100644 --- a/plugins/google-genai/src/main/java/com/google/genkit/plugins/googlegenai/GeminiModel.java +++ b/plugins/google-genai/src/main/java/com/google/genkit/plugins/googlegenai/GeminiModel.java @@ -63,7 +63,7 @@ public class GeminiModel implements Model { /** * Creates a new GeminiModel. * - * @param modelName the model name (e.g., "gemini-2.0-flash", "gemini-2.5-pro") + * @param modelName the model name (e.g., "gemini-2.5-flash", "gemini-2.5-pro") * @param options the plugin options */ public GeminiModel(String modelName, GoogleGenAIPluginOptions options) { diff --git a/plugins/mongodb/pom.xml b/plugins/mongodb/pom.xml new file mode 100644 index 000000000..d7e8dba19 --- /dev/null +++ b/plugins/mongodb/pom.xml @@ -0,0 +1,80 @@ + + + + 4.0.0 + + + com.google.genkit + genkit-parent + 1.0.0-SNAPSHOT + ../../pom.xml + + + genkit-plugin-mongodb + jar + Genkit MongoDB Plugin + MongoDB integration for Genkit - agent session persistence backed by MongoDB + + + false + 5.8.0 + + + + + + com.google.genkit + genkit-core + ${project.version} + + + com.google.genkit + genkit-ai + ${project.version} + + + + + org.mongodb + mongodb-driver-sync + ${mongodb.version} + + + + + org.slf4j + slf4j-api + + + + + org.junit.jupiter + junit-jupiter + test + + + org.mockito + mockito-core + test + + + diff --git a/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/MongoPlugin.java b/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/MongoPlugin.java new file mode 100644 index 000000000..accf1f4e0 --- /dev/null +++ b/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/MongoPlugin.java @@ -0,0 +1,183 @@ +/* + * Copyright 2025 Google LLC + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * + * SPDX-License-Identifier: Apache-2.0 + */ + +package com.google.genkit.plugins.mongodb; + +import com.google.genkit.ai.Embedder; +import com.google.genkit.core.Action; +import com.google.genkit.core.ActionType; +import com.google.genkit.core.Plugin; +import com.google.genkit.core.Registry; +import com.mongodb.client.MongoClient; +import com.mongodb.client.MongoClients; +import java.util.ArrayList; +import java.util.List; +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + +/** + * MongoDB Atlas Vector Search plugin for Genkit. + * + *

Registers a retriever and indexer named {@code mongodb/} for each configured + * collection, backed by an Atlas Vector Search index. Requires MongoDB Atlas or the {@code + * mongodb/mongodb-atlas-local} Docker image (a plain MongoDB server does not support {@code + * $vectorSearch}). + * + *

Example usage: + * + * 

{@code
+ * Genkit genkit = Genkit.builder()
+ *     .plugin(GoogleGenAIPlugin.create(apiKey))
+ *     .plugin(
+ *         MongoPlugin.builder()
+ *             .connectionString("mongodb://localhost:27017/?directConnection=true")
+ *             .addCollection(
+ *                 MongoVectorStoreConfig.builder()
+ *                     .collectionName("films")
+ *                     .embedderName("googleai/gemini-embedding-001")
+ *                     .dimension(768)
+ *                     .createIndexIfNotExists(true)
+ *                     .build())
+ *             .build())
+ *     .build();
+ * }
+ */ +public final class MongoPlugin implements Plugin { + + /** The plugin name; used as the {@code mongodb/...} action prefix. */ + public static final String PLUGIN_NAME = "mongodb"; + + private static final Logger logger = LoggerFactory.getLogger(MongoPlugin.class); + + private final String connectionString; + private final MongoClient externalClient; + private final List collectionConfigs; + private MongoClient client; + + private MongoPlugin(Builder builder) { + this.connectionString = builder.connectionString; + this.externalClient = builder.externalClient; + this.collectionConfigs = new ArrayList<>(builder.collectionConfigs); + } + + /** + * Creates a new builder. + * + * @return a new builder + */ + public static Builder builder() { + return new Builder(); + } + + @Override + public String getName() { + return PLUGIN_NAME; + } + + @Override + public List> init() { + throw new IllegalStateException( + "MongoPlugin requires a Registry to resolve embedders. Use init(registry) instead."); + } + + @Override + public List> init(Registry registry) { + client = externalClient != null ? externalClient : MongoClients.create(connectionString); + + List> actions = new ArrayList<>(); + for (MongoVectorStoreConfig config : collectionConfigs) { + String embedderKey = ActionType.EMBEDDER.keyFromName(config.getEmbedderName()); + Action embedderAction = registry.lookupAction(embedderKey); + if (embedderAction == null) { + throw new IllegalStateException( + "Embedder not found: " + + config.getEmbedderName() + + ". Make sure the embedder plugin is registered before MongoPlugin."); + } + if (!(embedderAction instanceof Embedder embedder)) { + throw new IllegalStateException( + "Action " + config.getEmbedderName() + " is not an Embedder"); + } + + MongoVectorStore store = new MongoVectorStore(client, config, embedder); + actions.add(store.createRetriever()); + actions.add(store.createIndexer()); + logger.info( + "Registered MongoDB vector store: {}/{}", PLUGIN_NAME, config.getCollectionName()); + } + return actions; + } + + /** Builder for {@link MongoPlugin}. */ + public static final class Builder { + private String connectionString; + private MongoClient externalClient; + private final List collectionConfigs = new ArrayList<>(); + + private Builder() {} + + /** + * Sets the MongoDB connection string (required unless an external client is provided). + * + * @param connectionString the connection string + * @return this builder + */ + public Builder connectionString(String connectionString) { + this.connectionString = connectionString; + return this; + } + + /** + * Sets an external MongoDB client to use instead of creating one from a connection string. + * + * @param client the client + * @return this builder + */ + public Builder client(MongoClient client) { + this.externalClient = client; + return this; + } + + /** + * Adds a collection configuration. + * + * @param config the collection configuration + * @return this builder + */ + public Builder addCollection(MongoVectorStoreConfig config) { + this.collectionConfigs.add(config); + return this; + } + + /** + * Builds the plugin. + * + * @return a new {@code MongoPlugin} + */ + public MongoPlugin build() { + if (externalClient == null && (connectionString == null || connectionString.isBlank())) { + throw new IllegalStateException( + "connectionString is required when not providing an external MongoClient"); + } + if (collectionConfigs.isEmpty()) { + throw new IllegalStateException("At least one collection configuration is required"); + } + return new MongoPlugin(this); + } + } +} diff --git a/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/MongoVectorStore.java b/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/MongoVectorStore.java new file mode 100644 index 000000000..431db234f --- /dev/null +++ b/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/MongoVectorStore.java @@ -0,0 +1,363 @@ +/* + * Copyright 2025 Google LLC + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * + * SPDX-License-Identifier: Apache-2.0 + */ + +package com.google.genkit.plugins.mongodb; + +import com.google.genkit.ai.Document; +import com.google.genkit.ai.EmbedRequest; +import com.google.genkit.ai.EmbedResponse; +import com.google.genkit.ai.Embedder; +import com.google.genkit.ai.Indexer; +import com.google.genkit.ai.IndexerRequest; +import com.google.genkit.ai.IndexerResponse; +import com.google.genkit.ai.Retriever; +import com.google.genkit.ai.RetrieverRequest; +import com.google.genkit.ai.RetrieverResponse; +import com.google.genkit.core.ActionContext; +import com.mongodb.MongoCommandException; +import com.mongodb.client.MongoClient; +import com.mongodb.client.MongoCollection; +import com.mongodb.client.MongoDatabase; +import com.mongodb.client.model.Filters; +import com.mongodb.client.model.ReplaceOptions; +import com.mongodb.client.model.SearchIndexModel; +import com.mongodb.client.model.SearchIndexType; +import java.util.ArrayList; +import java.util.HashMap; +import java.util.List; +import java.util.Map; +import java.util.UUID; +import org.bson.conversions.Bson; +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + +/** + * MongoDB Atlas Vector Search-backed vector store. + * + *

Indexes documents into a collection with an embedding field and retrieves them with the {@code + * $vectorSearch} aggregation stage. Requires an Atlas Vector Search index; use the {@code + * mongodb/mongodb-atlas-local} Docker image for local development or a MongoDB Atlas cluster. + */ +public final class MongoVectorStore { + + private static final Logger logger = LoggerFactory.getLogger(MongoVectorStore.class); + private static final String SCORE_FIELD = "__score"; + private static final int INDEX_READY_TIMEOUT_SECONDS = 120; + + private final MongoCollection collection; + private final MongoVectorStoreConfig config; + private final Embedder embedder; + private final MongoDatabase database; + private boolean initialized = false; + + /** + * Creates a new store. + * + * @param client the MongoDB client + * @param config the collection configuration + * @param embedder the embedder used to vectorize documents and queries + */ + public MongoVectorStore(MongoClient client, MongoVectorStoreConfig config, Embedder embedder) { + this.config = config; + this.embedder = embedder; + this.database = client.getDatabase(config.getDatabaseName()); + this.collection = database.getCollection(config.getCollectionName()); + } + + /** Creates the retriever action registered by the plugin. */ + Retriever createRetriever() { + String name = MongoPlugin.PLUGIN_NAME + "/" + config.getCollectionName(); + return Retriever.builder().name(name).handler(this::retrieve).build(); + } + + /** Creates the indexer action registered by the plugin. */ + Indexer createIndexer() { + String name = MongoPlugin.PLUGIN_NAME + "/" + config.getCollectionName(); + return Indexer.builder().name(name).handler(this::index).build(); + } + + private synchronized void ensureInitialized() { + if (initialized) { + return; + } + if (config.isCreateIndexIfNotExists()) { + ensureCollection(); + ensureVectorIndex(); + } + initialized = true; + logger.info("MongoDB vector store initialized for collection: {}", config.getCollectionName()); + } + + private void ensureCollection() { + boolean exists = false; + for (String name : database.listCollectionNames()) { + if (name.equals(config.getCollectionName())) { + exists = true; + break; + } + } + if (!exists) { + database.createCollection(config.getCollectionName()); + } + } + + private void ensureVectorIndex() { + if (!searchIndexExists()) { + Bson definition = + new org.bson.Document( + "fields", + List.of( + new org.bson.Document("type", "vector") + .append("path", config.getEmbeddingField()) + .append("numDimensions", resolveDimension()) + .append("similarity", config.getSimilarity().getValue()))); + createSearchIndexWithRetry(definition); + logger.info("Creating Atlas Vector Search index: {}", config.getIndexName()); + } + waitForIndexReady(); + } + + /** + * Resolves the embedding dimension by probing the embedder, falling back to the configured + * dimension if the probe fails. This keeps the created index in sync with whatever embedding + * model is wired in. + */ + private int resolveDimension() { + try { + return generateEmbedding(null, "genkit dimension probe").size(); + } catch (RuntimeException e) { + logger.debug( + "Embedding probe failed; using configured dimension {}: {}", + config.getDimension(), + e.getMessage()); + return config.getDimension(); + } + } + + /** + * Returns whether the configured vector index already exists, retrying while the Atlas Search + * service ({@code mongot}) is still starting up (error 125). The service can lag behind {@code + * mongod} readiness, especially with the local Atlas image. + */ + private boolean searchIndexExists() { + for (org.bson.Document idx : listSearchIndexesWithRetry()) { + if (config.getIndexName().equals(idx.getString("name"))) { + return true; + } + } + return false; + } + + private void createSearchIndexWithRetry(Bson definition) { + long deadline = System.currentTimeMillis() + INDEX_READY_TIMEOUT_SECONDS * 1000L; + while (true) { + try { + collection.createSearchIndexes( + List.of( + new SearchIndexModel( + config.getIndexName(), definition, SearchIndexType.vectorSearch()))); + return; + } catch (MongoCommandException e) { + if (isSearchServiceUnavailable(e) && System.currentTimeMillis() < deadline) { + logger.info("Waiting for Atlas Search service before creating the vector index..."); + sleep(2000); + continue; + } + throw e; + } + } + } + + private List listSearchIndexesWithRetry() { + long deadline = System.currentTimeMillis() + INDEX_READY_TIMEOUT_SECONDS * 1000L; + while (true) { + try { + List out = new java.util.ArrayList<>(); + for (org.bson.Document idx : collection.listSearchIndexes()) { + out.add(idx); + } + return out; + } catch (MongoCommandException e) { + if (isSearchServiceUnavailable(e) && System.currentTimeMillis() < deadline) { + logger.info("Waiting for Atlas Search service to become available..."); + sleep(2000); + continue; + } + throw e; + } + } + } + + private static boolean isSearchServiceUnavailable(MongoCommandException e) { + return e.getErrorCode() == 125 + || (e.getErrorMessage() != null + && e.getErrorMessage().contains("Search Index Management service")); + } + + private static void sleep(long millis) { + try { + Thread.sleep(millis); + } catch (InterruptedException e) { + Thread.currentThread().interrupt(); + throw new RuntimeException("Interrupted while waiting for the Atlas Search service", e); + } + } + + private void waitForIndexReady() { + long deadline = System.currentTimeMillis() + INDEX_READY_TIMEOUT_SECONDS * 1000L; + while (System.currentTimeMillis() < deadline) { + for (org.bson.Document idx : listSearchIndexesWithRetry()) { + if (config.getIndexName().equals(idx.getString("name")) + && Boolean.TRUE.equals(idx.getBoolean("queryable"))) { + logger.info("Atlas Vector Search index {} is queryable", config.getIndexName()); + return; + } + } + sleep(2000); + } + throw new RuntimeException( + "Timed out waiting for Atlas Vector Search index " + config.getIndexName()); + } + + /** + * Retrieves documents similar to the query using the {@code $vectorSearch} aggregation stage. + * + * @param context the action context + * @param request the retriever request + * @return the retriever response with matching documents + */ + public RetrieverResponse retrieve(ActionContext context, RetrieverRequest request) { + ensureInitialized(); + Document queryDoc = request.getQuery(); + if (queryDoc == null || queryDoc.text() == null || queryDoc.text().isBlank()) { + throw new RuntimeException("Query document has no text content"); + } + int topK = + request.getOptions() != null && request.getOptions().getK() != null + ? request.getOptions().getK() + : 10; + List queryVector = generateEmbedding(context, queryDoc.text()); + int numCandidates = Math.max(config.getNumCandidates(), topK * 10); + + List pipeline = + List.of( + new org.bson.Document( + "$vectorSearch", + new org.bson.Document("index", config.getIndexName()) + .append("path", config.getEmbeddingField()) + .append("queryVector", queryVector) + .append("numCandidates", numCandidates) + .append("limit", topK)), + new org.bson.Document( + "$addFields", + new org.bson.Document( + SCORE_FIELD, new org.bson.Document("$meta", "vectorSearchScore")))); + + List documents = new ArrayList<>(); + for (org.bson.Document result : collection.aggregate(pipeline)) { + documents.add(toDocument(result)); + } + logger.debug( + "Retrieved {} documents from collection {}", documents.size(), config.getCollectionName()); + return new RetrieverResponse(documents); + } + + /** + * Indexes documents into the collection, generating an embedding for each. + * + * @param context the action context + * @param request the indexer request + * @return the indexer response + */ + public IndexerResponse index(ActionContext context, IndexerRequest request) { + ensureInitialized(); + List documents = request.getDocuments(); + if (documents == null || documents.isEmpty()) { + logger.warn("No documents to index"); + return new IndexerResponse(); + } + int indexed = 0; + for (Document doc : documents) { + String content = doc.text() != null ? doc.text() : ""; + List embedding = generateEmbedding(context, content); + String id = getOrGenerateId(doc); + + org.bson.Document stored = new org.bson.Document("_id", id); + stored.put(config.getTextField(), content); + stored.put(config.getEmbeddingField(), embedding); + if (doc.getMetadata() != null) { + for (Map.Entry entry : doc.getMetadata().entrySet()) { + if (!"id".equals(entry.getKey())) { + stored.put(entry.getKey(), entry.getValue()); + } + } + } + for (Map.Entry entry : config.getAdditionalMetadata().entrySet()) { + stored.put(entry.getKey(), entry.getValue()); + } + collection.replaceOne(Filters.eq("_id", id), stored, new ReplaceOptions().upsert(true)); + indexed++; + } + logger.info("Indexed {} documents into collection {}", indexed, config.getCollectionName()); + return new IndexerResponse(); + } + + private Document toDocument(org.bson.Document result) { + Map metadata = new HashMap<>(); + String content = ""; + for (Map.Entry entry : result.entrySet()) { + String key = entry.getKey(); + if (key.equals(config.getTextField())) { + content = entry.getValue() != null ? entry.getValue().toString() : ""; + } else if (key.equals(config.getEmbeddingField()) || key.equals(SCORE_FIELD)) { + continue; + } else if (key.equals("_id")) { + metadata.put("id", entry.getValue() != null ? entry.getValue().toString() : null); + } else { + metadata.put(key, entry.getValue()); + } + } + Object score = result.get(SCORE_FIELD); + if (score instanceof Number number) { + metadata.put("score", number.doubleValue()); + } + Document doc = new Document(content); + doc.setMetadata(metadata); + return doc; + } + + private List generateEmbedding(ActionContext ctx, String text) { + EmbedResponse response = embedder.run(ctx, new EmbedRequest(List.of(new Document(text)))); + if (response.getEmbeddings() == null || response.getEmbeddings().isEmpty()) { + throw new RuntimeException("Failed to generate embedding for text"); + } + float[] values = response.getEmbeddings().get(0).getValues(); + List out = new ArrayList<>(values.length); + for (float v : values) { + out.add((double) v); + } + return out; + } + + private String getOrGenerateId(Document doc) { + if (doc.getMetadata() != null && doc.getMetadata().get("id") != null) { + return doc.getMetadata().get("id").toString(); + } + return UUID.randomUUID().toString(); + } +} diff --git a/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/MongoVectorStoreConfig.java b/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/MongoVectorStoreConfig.java new file mode 100644 index 000000000..41f937457 --- /dev/null +++ b/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/MongoVectorStoreConfig.java @@ -0,0 +1,355 @@ +/* + * Copyright 2025 Google LLC + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * + * SPDX-License-Identifier: Apache-2.0 + */ + +package com.google.genkit.plugins.mongodb; + +import java.util.HashMap; +import java.util.Map; + +/** + * Configuration for a single MongoDB Atlas Vector Search collection managed by {@link MongoPlugin}. + * + *

Each config registers a retriever and indexer named {@code mongodb/} backed by + * an Atlas Vector Search index over the {@link #getEmbeddingField() embedding field}. + */ +public final class MongoVectorStoreConfig { + + /** Vector similarity function supported by Atlas Vector Search. */ + public enum Similarity { + COSINE("cosine"), + EUCLIDEAN("euclidean"), + DOT_PRODUCT("dotProduct"); + + private final String value; + + Similarity(String value) { + this.value = value; + } + + /** + * Returns the Atlas Vector Search similarity name. + * + * @return the similarity name + */ + public String getValue() { + return value; + } + } + + private final String databaseName; + private final String collectionName; + private final String embedderName; + private final String indexName; + private final int dimension; + private final Similarity similarity; + private final String textField; + private final String embeddingField; + private final int numCandidates; + private final boolean createIndexIfNotExists; + private final Map additionalMetadata; + + private MongoVectorStoreConfig(Builder builder) { + this.databaseName = builder.databaseName; + this.collectionName = builder.collectionName; + this.embedderName = builder.embedderName; + this.indexName = builder.indexName; + this.dimension = builder.dimension; + this.similarity = builder.similarity; + this.textField = builder.textField; + this.embeddingField = builder.embeddingField; + this.numCandidates = builder.numCandidates; + this.createIndexIfNotExists = builder.createIndexIfNotExists; + this.additionalMetadata = new HashMap<>(builder.additionalMetadata); + } + + /** + * Returns the database name (default {@code genkit}). + * + * @return the database name + */ + public String getDatabaseName() { + return databaseName; + } + + /** + * Returns the collection name. + * + * @return the collection name + */ + public String getCollectionName() { + return collectionName; + } + + /** + * Returns the name of the embedder used to vectorize documents and queries. + * + * @return the embedder name + */ + public String getEmbedderName() { + return embedderName; + } + + /** + * Returns the Atlas Vector Search index name (default {@code genkit_vector_index}). + * + * @return the index name + */ + public String getIndexName() { + return indexName; + } + + /** + * Returns the embedding dimension (default {@code 768}). + * + * @return the embedding dimension + */ + public int getDimension() { + return dimension; + } + + /** + * Returns the vector similarity function (default {@link Similarity#COSINE}). + * + * @return the similarity + */ + public Similarity getSimilarity() { + return similarity; + } + + /** + * Returns the field that stores the document text (default {@code text}). + * + * @return the text field name + */ + public String getTextField() { + return textField; + } + + /** + * Returns the field that stores the embedding vector (default {@code embedding}). + * + * @return the embedding field name + */ + public String getEmbeddingField() { + return embeddingField; + } + + /** + * Returns the number of nearest neighbors to consider during the vector search (default {@code + * 100}). Atlas recommends a value at least 10× the requested result count. + * + * @return the number of candidates + */ + public int getNumCandidates() { + return numCandidates; + } + + /** + * Returns whether to create the Atlas Vector Search index on first use if it does not exist + * (default {@code false}). + * + * @return {@code true} if the index should be created when missing + */ + public boolean isCreateIndexIfNotExists() { + return createIndexIfNotExists; + } + + /** + * Returns additional metadata merged into every indexed document. + * + * @return the additional metadata + */ + public Map getAdditionalMetadata() { + return additionalMetadata; + } + + /** + * Creates a new builder. + * + * @return a new builder + */ + public static Builder builder() { + return new Builder(); + } + + /** Builder for {@link MongoVectorStoreConfig}. */ + public static final class Builder { + private String databaseName = "genkit"; + private String collectionName; + private String embedderName; + private String indexName = "genkit_vector_index"; + private int dimension = 768; + private Similarity similarity = Similarity.COSINE; + private String textField = "text"; + private String embeddingField = "embedding"; + private int numCandidates = 100; + private boolean createIndexIfNotExists = false; + private final Map additionalMetadata = new HashMap<>(); + + private Builder() {} + + /** + * Sets the database name. + * + * @param databaseName the database name + * @return this builder + */ + public Builder databaseName(String databaseName) { + this.databaseName = databaseName; + return this; + } + + /** + * Sets the collection name. + * + * @param collectionName the collection name + * @return this builder + */ + public Builder collectionName(String collectionName) { + this.collectionName = collectionName; + return this; + } + + /** + * Sets the embedder name. + * + * @param embedderName the embedder name + * @return this builder + */ + public Builder embedderName(String embedderName) { + this.embedderName = embedderName; + return this; + } + + /** + * Sets the Atlas Vector Search index name. + * + * @param indexName the index name + * @return this builder + */ + public Builder indexName(String indexName) { + this.indexName = indexName; + return this; + } + + /** + * Sets the embedding dimension. + * + * @param dimension the embedding dimension (must be {@code >= 1}) + * @return this builder + */ + public Builder dimension(int dimension) { + this.dimension = dimension; + return this; + } + + /** + * Sets the vector similarity function. + * + * @param similarity the similarity + * @return this builder + */ + public Builder similarity(Similarity similarity) { + this.similarity = similarity; + return this; + } + + /** + * Sets the text field name. + * + * @param textField the text field name + * @return this builder + */ + public Builder textField(String textField) { + this.textField = textField; + return this; + } + + /** + * Sets the embedding field name. + * + * @param embeddingField the embedding field name + * @return this builder + */ + public Builder embeddingField(String embeddingField) { + this.embeddingField = embeddingField; + return this; + } + + /** + * Sets the number of nearest neighbors to consider during the vector search. + * + * @param numCandidates the number of candidates (must be {@code >= 1}) + * @return this builder + */ + public Builder numCandidates(int numCandidates) { + this.numCandidates = numCandidates; + return this; + } + + /** + * Sets whether to create the Atlas Vector Search index on first use if it does not exist. + * + * @param createIndexIfNotExists whether to create the index when missing + * @return this builder + */ + public Builder createIndexIfNotExists(boolean createIndexIfNotExists) { + this.createIndexIfNotExists = createIndexIfNotExists; + return this; + } + + /** + * Adds a metadata entry merged into every indexed document. + * + * @param key the metadata key + * @param value the metadata value + * @return this builder + */ + public Builder addAdditionalMetadata(String key, Object value) { + this.additionalMetadata.put(key, value); + return this; + } + + /** + * Builds a new {@code MongoVectorStoreConfig}. + * + * @return a new config instance + */ + public MongoVectorStoreConfig build() { + if (databaseName == null || databaseName.isBlank()) { + throw new IllegalArgumentException("databaseName must be non-empty"); + } + if (collectionName == null || collectionName.isBlank()) { + throw new IllegalArgumentException("collectionName must be non-empty"); + } + if (embedderName == null || embedderName.isBlank()) { + throw new IllegalArgumentException("embedderName must be non-empty"); + } + if (indexName == null || indexName.isBlank()) { + throw new IllegalArgumentException("indexName must be non-empty"); + } + if (dimension < 1) { + throw new IllegalArgumentException("dimension must be >= 1"); + } + if (numCandidates < 1) { + throw new IllegalArgumentException("numCandidates must be >= 1"); + } + return new MongoVectorStoreConfig(this); + } + } +} diff --git a/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/package-info.java b/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/package-info.java new file mode 100644 index 000000000..f73db19df --- /dev/null +++ b/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/package-info.java @@ -0,0 +1,26 @@ +/* + * Copyright 2025 Google LLC + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * + * SPDX-License-Identifier: Apache-2.0 + */ + +/** + * MongoDB integration for Genkit. + * + *

{@link com.google.genkit.plugins.mongodb.MongoPlugin} registers MongoDB Atlas Vector Search + * retrievers and indexers for RAG workflows. Agent session persistence lives in the {@link + * com.google.genkit.plugins.mongodb.session} subpackage. + */ +package com.google.genkit.plugins.mongodb; diff --git a/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/session/MongoSessionStore.java b/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/session/MongoSessionStore.java new file mode 100644 index 000000000..be788130e --- /dev/null +++ b/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/session/MongoSessionStore.java @@ -0,0 +1,723 @@ +/* + * Copyright 2025 Google LLC + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * + * SPDX-License-Identifier: Apache-2.0 + */ + +package com.google.genkit.plugins.mongodb.session; + +import com.fasterxml.jackson.databind.JsonNode; +import com.fasterxml.jackson.databind.ObjectMapper; +import com.fasterxml.jackson.databind.node.ArrayNode; +import com.fasterxml.jackson.databind.node.NullNode; +import com.fasterxml.jackson.databind.node.ObjectNode; +import com.google.genkit.ai.agent.AgentFinishReason; +import com.google.genkit.ai.agent.GetSnapshotOptions; +import com.google.genkit.ai.agent.RuntimeError; +import com.google.genkit.ai.agent.SessionSnapshot; +import com.google.genkit.ai.agent.SessionState; +import com.google.genkit.ai.agent.SessionStore; +import com.google.genkit.ai.agent.SessionStoreOptions; +import com.google.genkit.ai.agent.SnapshotMutator; +import com.google.genkit.ai.agent.SnapshotStatus; +import com.google.genkit.ai.agent.SnapshotSubscriber; +import com.google.genkit.ai.agent.internal.SnapshotSharding; +import com.google.genkit.core.GenkitException; +import com.google.genkit.core.JsonUtils; +import com.google.genkit.core.jsonpatch.JsonPatch; +import com.mongodb.ErrorCategory; +import com.mongodb.MongoWriteException; +import com.mongodb.client.MongoClient; +import com.mongodb.client.MongoCollection; +import com.mongodb.client.model.Filters; +import com.mongodb.client.model.ReplaceOptions; +import com.mongodb.client.result.UpdateResult; +import java.nio.charset.StandardCharsets; +import java.util.ArrayList; +import java.util.List; +import java.util.UUID; +import java.util.concurrent.Executors; +import java.util.concurrent.ScheduledExecutorService; +import java.util.concurrent.ScheduledFuture; +import java.util.concurrent.TimeUnit; +import java.util.function.Consumer; +import org.bson.Document; +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + +/** + * MongoDB-backed implementation of {@link SessionStore} and {@link SnapshotSubscriber}. + * + *

Persists session snapshots with the same sharded checkpoint + diff + pointer layout as the + * Firestore, DynamoDB, Cosmos DB, and PostgreSQL backends (see {@code FirestoreSessionStore}), + * sharing the pure logic in {@link SnapshotSharding}. + * + *

Storage layout (single collection)

+ * + *

All documents live in one collection (default database {@code genkit}, collection {@code + * genkit_sessions}). Each document's {@code _id} is {@code ::} where {@code + * prefix} is the per-tenant prefix (default {@code "global"}) and {@code recordId} discriminates + * the record kind: + * + *

    + *
  • {@code SNAP_} — one metadata document per snapshot. {@code kind} is {@code + * "checkpoint"} or {@code "diff"}; carries {@code checkpointId}, {@code + * checkpointShardCount}, {@code segmentPath}, {@code statePatch} (RFC-6902 as a JSON string + * for diffs) and {@code error} (JSON string). + *
  • {@code SHARD__} — a shard of the checkpoint state JSON. + *
  • {@code PTR_} — the current leaf pointer for a session. + *
+ * + *

Concurrency

+ * + *

Shard and snapshot documents are idempotent by {@code _id} (upserted); updating an existing + * snapshot id uses a {@code version} conditional replace, re-applying the (pure) mutator on + * conflict. The session pointer is advanced monotonically (never backward) under the same {@code + * version} concurrency. Shards are written before the snapshot document, which is written before + * the pointer flips, so a reader following the pointer always sees complete data. + * + * @param the type of custom session state + */ +public final class MongoSessionStore implements SessionStore, SnapshotSubscriber { + + private static final Logger logger = LoggerFactory.getLogger(MongoSessionStore.class); + private static final ObjectMapper MAPPER = JsonUtils.getObjectMapper(); + + static final String KIND_CHECKPOINT = "checkpoint"; + static final String KIND_DIFF = "diff"; + + private static final int MAX_ATTEMPTS = 5; + + private final MongoCollection collection; + private final MongoSessionStoreOptions options; + private final ScheduledExecutorService scheduler; + + /** + * Creates a store with default options. + * + * @param client the MongoDB client + */ + public MongoSessionStore(MongoClient client) { + this(client, MongoSessionStoreOptions.defaults()); + } + + /** + * Creates a store. + * + * @param client the MongoDB client + * @param options the store options + */ + public MongoSessionStore(MongoClient client, MongoSessionStoreOptions options) { + if (client == null) { + throw new IllegalArgumentException("MongoClient must be non-null"); + } + if (options == null) { + throw new IllegalArgumentException("options must be non-null"); + } + this.options = options; + this.collection = + client.getDatabase(options.getDatabaseName()).getCollection(options.getCollectionName()); + this.scheduler = + Executors.newSingleThreadScheduledExecutor( + r -> { + Thread t = new Thread(r, "genkit-mongo-session-store-poll"); + t.setDaemon(true); + return t; + }); + } + + // ────────────────────────────────────────────────────────────────────────── + // Id helpers + // ────────────────────────────────────────────────────────────────────────── + + private String prefix(SessionStoreOptions opts) { + String p = + options.getSnapshotPathPrefix().apply(opts != null ? opts : SessionStoreOptions.empty()); + return (p == null || p.isBlank()) ? "global" : p; + } + + private static String key(String prefix, String recordId) { + return prefix + "::" + recordId; + } + + private static String snapId(String snapshotId) { + return "SNAP_" + snapshotId; + } + + private static String shardId(String checkpointId, int index) { + return "SHARD_" + checkpointId + "_" + index; + } + + private static String ptrId(String sessionId) { + return "PTR_" + sessionId; + } + + // ────────────────────────────────────────────────────────────────────────── + // SnapshotWriter + // ────────────────────────────────────────────────────────────────────────── + + /** + * {@inheritDoc} + * + *

Implements the same identity/sessionId/status defaulting contract as the reference stores, + * then writes the snapshot (checkpoint shards + metadata document) and advances the session + * pointer. + */ + @Override + public String saveSnapshot( + String snapshotId, SnapshotMutator mutator, SessionStoreOptions storeOpts) { + String prefix = prefix(storeOpts); + + for (int attempt = 1; ; attempt++) { + // 1. Read existing snapshot (+ version) and reconstruct state. + SessionSnapshot existing = null; + String existingSessionId = null; + Long existingVersion = null; + if (snapshotId != null) { + SnapshotSharding.validateId(snapshotId); + Row row = readRow(prefix, snapId(snapshotId)); + if (row != null) { + existing = readSnapshot(prefix, row.doc); + existingSessionId = existing.getSessionId(); + existingVersion = row.version; + } + } + + // 2. Apply mutator (pure — safe to re-run on conflict retry). + SessionSnapshot result = mutator.apply(existing); + if (result == null) { + return null; + } + + // 3. Identity / sessionId / status defaulting (mirror the reference stores). + String finalId; + if (snapshotId != null) { + finalId = snapshotId; + } else if (result.getSnapshotId() != null && !result.getSnapshotId().isBlank()) { + finalId = result.getSnapshotId(); + } else { + finalId = UUID.randomUUID().toString(); + } + SnapshotSharding.validateId(finalId); + result.setSnapshotId(finalId); + + if (existingSessionId != null) { + result.setSessionId(existingSessionId); + } + if (result.getSessionId() == null && result.getState() != null) { + result.setSessionId(result.getState().getSessionId()); + } + if (result.getSessionId() == null || result.getSessionId().isBlank()) { + throw GenkitException.builder() + .message("snapshot requires sessionId") + .errorCode("INVALID_ARGUMENT") + .build(); + } + if (result.getStatus() == null) { + result.setStatus(SnapshotStatus.COMPLETED); + } + + // 4. Resolve parent metadata to decide checkpoint vs diff. + String parentId = result.getParentId(); + JsonNode newState = stateToJson(result.getState()); + + ParentInfo parent = null; + if (parentId != null && !parentId.isBlank()) { + Row parentRow = readRow(prefix, snapId(parentId)); + if (parentRow != null) { + parent = loadParentInfo(prefix, parentRow.doc); + } + } + + boolean parentExists = parent != null; + int depthFromCheckpoint = 0; + JsonNode statePatch = null; + int diffSizeBytes = 0; + if (parent != null) { + depthFromCheckpoint = parent.segmentPath.size() + 1; + statePatch = JsonPatch.diff(parent.state, newState); + diffSizeBytes = jsonBytes(statePatch); + } + + boolean checkpoint = + SnapshotSharding.shouldCheckpoint( + parentExists, + depthFromCheckpoint, + options.getCheckpointInterval(), + diffSizeBytes, + options.getShardSize()); + + // 5. Build the snapshot metadata document (+ shard docs for a checkpoint). + ObjectNode doc = baseDoc(result); + String checkpointId; + int checkpointShardCount; + List segmentPath; + + if (checkpoint) { + checkpointId = finalId; + String stateJson = writeJson(newState); + List shards = SnapshotSharding.shardString(stateJson, options.getShardSize()); + checkpointShardCount = shards.size(); + segmentPath = new ArrayList<>(); + for (int i = 0; i < shards.size(); i++) { + ObjectNode shardDoc = MAPPER.createObjectNode(); + shardDoc.put("checkpointId", checkpointId); + shardDoc.put("index", i); + shardDoc.put("data", shards.get(i)); + upsertRow(prefix, shardId(checkpointId, i), shardDoc); + } + doc.put("kind", KIND_CHECKPOINT); + } else { + ParentInfo p = parent; + if (p == null) { + throw new GenkitException("internal: diff path without parent"); + } + checkpointId = p.checkpointId; + checkpointShardCount = p.checkpointShardCount; + segmentPath = new ArrayList<>(p.segmentPath); + segmentPath.add(finalId); + doc.put("kind", KIND_DIFF); + doc.put("statePatch", writeJson(statePatch)); + } + + doc.put("checkpointId", checkpointId); + doc.put("checkpointShardCount", checkpointShardCount); + putStringArray(doc, "segmentPath", segmentPath); + + // 6. Write the snapshot document with optimistic concurrency. + boolean written = writeConditional(prefix, snapId(finalId), doc, existingVersion); + if (!written) { + if (attempt < MAX_ATTEMPTS) { + continue; // concurrent writer won the race; re-read and retry the pure mutator. + } + throw new GenkitException("Failed to save snapshot after " + MAX_ATTEMPTS + " attempts"); + } + + // 7. Advance the session pointer (never backward). + advancePointer( + prefix, + result.getSessionId(), + finalId, + result.getCreatedAt(), + result.getUpdatedAt(), + checkpointId, + checkpointShardCount, + segmentPath); + + return finalId; + } + } + + /** Advances the session pointer to the new leaf unless the stored pointer is already newer. */ + private void advancePointer( + String prefix, + String sessionId, + String snapshotId, + String createdAt, + String updatedAt, + String checkpointId, + int checkpointShardCount, + List segmentPath) { + String id = ptrId(sessionId); + + for (int attempt = 1; ; attempt++) { + Row existing = readRow(prefix, id); + Long version = existing != null ? existing.version : null; + if (existing != null) { + String currentLeaf = getString(existing.doc, "currentSnapshotId"); + String currentCreatedAt = getString(existing.doc, "currentCreatedAt"); + boolean sameLeaf = snapshotId.equals(currentLeaf); + boolean newer = + createdAt == null + || currentCreatedAt == null + || createdAt.compareTo(currentCreatedAt) >= 0; + if (!sameLeaf && !newer) { + return; // stored pointer is already newer; don't move backward. + } + } + + ObjectNode ptr = MAPPER.createObjectNode(); + ptr.put("currentSnapshotId", snapshotId); + ptr.put("checkpointId", checkpointId); + ptr.put("checkpointShardCount", checkpointShardCount); + putStringArray(ptr, "segmentPath", segmentPath); + if (createdAt != null) { + ptr.put("currentCreatedAt", createdAt); + } + if (updatedAt != null) { + ptr.put("updatedAt", updatedAt); + } + + if (writeConditional(prefix, id, ptr, version)) { + return; + } + if (attempt >= MAX_ATTEMPTS) { + logger.debug("Pointer for session {} not advanced (lost concurrency race)", sessionId); + return; + } + } + } + + // ────────────────────────────────────────────────────────────────────────── + // SnapshotReader + // ────────────────────────────────────────────────────────────────────────── + + /** + * {@inheritDoc} + * + *

By {@code snapshotId}: loads the snapshot document and reconstructs its state from the + * checkpoint shards + ordered {@code segmentPath} diffs. By {@code sessionId}: reads the pointer, + * then loads and reconstructs the pointed snapshot. + */ + @Override + public SessionSnapshot getSnapshot(GetSnapshotOptions opts) { + if (opts == null) { + return null; + } + String prefix = prefix(SessionStoreOptions.empty()); + + if (opts.getSnapshotId() != null) { + SnapshotSharding.validateId(opts.getSnapshotId()); + Row row = readRow(prefix, snapId(opts.getSnapshotId())); + return row == null ? null : readSnapshot(prefix, row.doc); + } + if (opts.getSessionId() != null) { + SnapshotSharding.validateId(opts.getSessionId()); + Row pointer = readRow(prefix, ptrId(opts.getSessionId())); + if (pointer == null) { + return null; + } + String leafId = getString(pointer.doc, "currentSnapshotId"); + if (leafId == null) { + return null; + } + Row row = readRow(prefix, snapId(leafId)); + return row == null ? null : readSnapshot(prefix, row.doc); + } + return null; + } + + // ────────────────────────────────────────────────────────────────────────── + // SnapshotSubscriber (polling) + // ────────────────────────────────────────────────────────────────────────── + + /** + * {@inheritDoc} + * + *

The subscription polls {@link #getSnapshot} on a shared daemon scheduler and fires the + * callback whenever the serialized snapshot content changes. The callback also fires immediately + * if the snapshot already exists. + */ + @Override + public AutoCloseable onSnapshotStateChange( + String snapshotId, Consumer> cb, SessionStoreOptions storeOpts) { + SnapshotSharding.validateId(snapshotId); + GetSnapshotOptions get = GetSnapshotOptions.builder().snapshotId(snapshotId).build(); + + final String[] lastContent = {null}; + SessionSnapshot initial = getSnapshot(get); + if (initial != null) { + lastContent[0] = serializeQuietly(initial); + cb.accept(initial); + } + + ScheduledFuture future = + scheduler.scheduleAtFixedRate( + () -> { + try { + SessionSnapshot snap = getSnapshot(get); + if (snap == null) { + return; + } + String content = serializeQuietly(snap); + if (!content.equals(lastContent[0])) { + lastContent[0] = content; + cb.accept(snap); + } + } catch (Exception e) { + // Swallow poll errors — don't kill the scheduler thread. + } + }, + options.getPollIntervalMs(), + options.getPollIntervalMs(), + TimeUnit.MILLISECONDS); + + return () -> future.cancel(false); + } + + // ────────────────────────────────────────────────────────────────────────── + // Document (de)serialization + // ────────────────────────────────────────────────────────────────────────── + + /** Builds the non-state base document for a snapshot (metadata only). */ + private ObjectNode baseDoc(SessionSnapshot snap) { + ObjectNode data = MAPPER.createObjectNode(); + data.put("snapshotId", snap.getSnapshotId()); + data.put("sessionId", snap.getSessionId()); + if (snap.getParentId() != null) { + data.put("parentId", snap.getParentId()); + } + if (snap.getCreatedAt() != null) { + data.put("createdAt", snap.getCreatedAt()); + } + if (snap.getUpdatedAt() != null) { + data.put("updatedAt", snap.getUpdatedAt()); + } + if (snap.getHeartbeatAt() != null) { + data.put("heartbeatAt", snap.getHeartbeatAt()); + } + if (snap.getStatus() != null) { + data.put("status", snap.getStatus().getValue()); + } + if (snap.getFinishReason() != null) { + data.put("finishReason", snap.getFinishReason().getValue()); + } + if (snap.getError() != null) { + data.put("error", writeJson(snap.getError())); + } + return data; + } + + /** Holds the reconstructed parent state and its checkpoint lineage. */ + private static final class ParentInfo { + JsonNode state; + String checkpointId; + int checkpointShardCount; + List segmentPath; + } + + /** Loads a parent snapshot's checkpoint lineage and reconstructs its state. */ + private ParentInfo loadParentInfo(String prefix, ObjectNode doc) { + ParentInfo info = new ParentInfo(); + info.checkpointId = getString(doc, "checkpointId"); + info.checkpointShardCount = getInt(doc, "checkpointShardCount"); + info.segmentPath = getStringList(doc, "segmentPath"); + info.state = + reconstructFullState( + prefix, info.checkpointId, info.checkpointShardCount, info.segmentPath); + return info; + } + + /** Reads and fully reconstructs a snapshot from its metadata document. */ + private SessionSnapshot readSnapshot(String prefix, ObjectNode doc) { + String checkpointId = getString(doc, "checkpointId"); + int checkpointShardCount = getInt(doc, "checkpointShardCount"); + List segmentPath = getStringList(doc, "segmentPath"); + JsonNode state = reconstructFullState(prefix, checkpointId, checkpointShardCount, segmentPath); + return docToSnapshot(doc, state); + } + + /** + * Reconstructs full state: loads the checkpoint shards (concatenate, parse) then applies the + * {@code segmentPath} diffs in order. + */ + private JsonNode reconstructFullState( + String prefix, String checkpointId, int checkpointShardCount, List segmentPath) { + if (checkpointId == null) { + return NullNode.getInstance(); + } + List shardContents = new ArrayList<>(); + for (int i = 0; i < checkpointShardCount; i++) { + Row shard = readRow(prefix, shardId(checkpointId, i)); + shardContents.add(shard != null ? getString(shard.doc, "data") : ""); + } + String checkpointJson = SnapshotSharding.reassembleShards(shardContents); + + List diffs = new ArrayList<>(); + for (String diffId : segmentPath) { + Row diffRow = readRow(prefix, snapId(diffId)); + if (diffRow != null) { + String patch = getString(diffRow.doc, "statePatch"); + if (patch != null) { + diffs.add(patch); + } + } + } + try { + return SnapshotSharding.reconstructState(checkpointJson, diffs); + } catch (Exception e) { + throw new GenkitException("Failed to reconstruct session state: " + e.getMessage(), e); + } + } + + /** Builds a {@link SessionSnapshot} from a metadata document and reconstructed state. */ + @SuppressWarnings("unchecked") + private SessionSnapshot docToSnapshot(ObjectNode doc, JsonNode state) { + SessionSnapshot.Builder builder = SessionSnapshot.builder(); + builder.snapshotId(getString(doc, "snapshotId")); + builder.sessionId(getString(doc, "sessionId")); + builder.parentId(getString(doc, "parentId")); + builder.createdAt(getString(doc, "createdAt")); + builder.updatedAt(getString(doc, "updatedAt")); + builder.heartbeatAt(getString(doc, "heartbeatAt")); + String status = getString(doc, "status"); + if (status != null) { + builder.status(SnapshotStatus.fromValueOrCompleted(status)); + } + String finishReason = getString(doc, "finishReason"); + if (finishReason != null) { + builder.finishReason(AgentFinishReason.fromValue(finishReason)); + } + String errorJson = getString(doc, "error"); + if (errorJson != null) { + try { + builder.error(MAPPER.readValue(errorJson, RuntimeError.class)); + } catch (Exception e) { + throw new GenkitException("Failed to parse snapshot error: " + e.getMessage(), e); + } + } + if (state != null && !state.isNull()) { + try { + builder.state((SessionState) MAPPER.treeToValue(state, SessionState.class)); + } catch (Exception e) { + throw new GenkitException("Failed to parse session state: " + e.getMessage(), e); + } + } + return builder.build(); + } + + /** Serializes session state to a JSON node (null state → JSON null). */ + private JsonNode stateToJson(SessionState state) { + if (state == null) { + return NullNode.getInstance(); + } + return MAPPER.valueToTree(state); + } + + // ────────────────────────────────────────────────────────────────────────── + // Low-level MongoDB + node helpers + // ────────────────────────────────────────────────────────────────────────── + + /** A single stored document: its JSON payload and optimistic-concurrency version. */ + private static final class Row { + final ObjectNode doc; + final long version; + + Row(ObjectNode doc, long version) { + this.doc = doc; + this.version = version; + } + } + + /** Reads a document by key, or {@code null} when it does not exist. */ + private Row readRow(String prefix, String id) { + Document found = collection.find(Filters.eq("_id", key(prefix, id))).first(); + if (found == null) { + return null; + } + long version = found.get("version") instanceof Number n ? n.longValue() : 1L; + Document payload = new Document(found); + payload.remove("_id"); + payload.remove("pk"); + payload.remove("version"); + try { + return new Row((ObjectNode) MAPPER.readTree(payload.toJson()), version); + } catch (Exception e) { + throw new GenkitException("Failed to read session document " + id + ": " + e.getMessage(), e); + } + } + + /** Builds the persisted document by merging the payload with the id/pk/version envelope. */ + private Document toDocument(String prefix, String id, ObjectNode payload, long version) { + Document doc = Document.parse(writeJson(payload)); + doc.put("_id", key(prefix, id)); + doc.put("pk", prefix); + doc.put("version", version); + return doc; + } + + /** Idempotently upserts a document (used for shard records). */ + private void upsertRow(String prefix, String id, ObjectNode payload) { + Document doc = toDocument(prefix, id, payload, 1L); + collection.replaceOne( + Filters.eq("_id", key(prefix, id)), doc, new ReplaceOptions().upsert(true)); + } + + /** + * Writes a document with optimistic concurrency: inserts when {@code expectedVersion} is {@code + * null}, otherwise replaces only when the stored version still matches. Returns {@code false} + * when a concurrent writer won the race (the caller re-reads and retries the pure mutator). + */ + private boolean writeConditional( + String prefix, String id, ObjectNode payload, Long expectedVersion) { + if (expectedVersion == null) { + Document doc = toDocument(prefix, id, payload, 1L); + try { + collection.insertOne(doc); + return true; + } catch (MongoWriteException e) { + if (e.getError().getCategory() == ErrorCategory.DUPLICATE_KEY) { + return false; // concurrent insert won the race. + } + throw new GenkitException( + "Failed to insert session document " + id + ": " + e.getMessage(), e); + } + } + Document doc = toDocument(prefix, id, payload, expectedVersion + 1); + UpdateResult result = + collection.replaceOne( + Filters.and(Filters.eq("_id", key(prefix, id)), Filters.eq("version", expectedVersion)), + doc); + return result.getMatchedCount() == 1; + } + + private static void putStringArray(ObjectNode node, String name, List values) { + ArrayNode arr = node.putArray(name); + for (String v : values) { + arr.add(v); + } + } + + private static String getString(ObjectNode node, String name) { + JsonNode v = node.get(name); + return (v == null || v.isNull()) ? null : v.asText(); + } + + private static int getInt(ObjectNode node, String name) { + JsonNode v = node.get(name); + return (v == null || v.isNull()) ? 0 : v.asInt(); + } + + private static List getStringList(ObjectNode node, String name) { + List out = new ArrayList<>(); + JsonNode v = node.get(name); + if (v != null && v.isArray()) { + for (JsonNode e : v) { + out.add(e.asText()); + } + } + return out; + } + + private static String writeJson(Object value) { + try { + return MAPPER.writeValueAsString(value); + } catch (Exception e) { + throw new GenkitException("Failed to serialize value: " + e.getMessage(), e); + } + } + + private static int jsonBytes(Object value) { + return writeJson(value).getBytes(StandardCharsets.UTF_8).length; + } + + private static String serializeQuietly(SessionSnapshot snap) { + try { + return MAPPER.writeValueAsString(MAPPER.valueToTree(snap)); + } catch (Exception e) { + return ""; + } + } +} diff --git a/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/session/MongoSessionStoreOptions.java b/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/session/MongoSessionStoreOptions.java new file mode 100644 index 000000000..c8edf97da --- /dev/null +++ b/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/session/MongoSessionStoreOptions.java @@ -0,0 +1,249 @@ +/* + * Copyright 2025 Google LLC + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * + * SPDX-License-Identifier: Apache-2.0 + */ + +package com.google.genkit.plugins.mongodb.session; + +import com.google.genkit.ai.agent.SessionStoreOptions; +import java.util.function.Function; + +/** + * Configuration for {@link MongoSessionStore}. + * + *

The store persists all records in a single MongoDB collection (default database {@value + * #DEFAULT_DATABASE}, collection {@value #DEFAULT_COLLECTION}). Each document's {@code _id} + * combines the per-tenant prefix (default {@code "global"}) with the record id; the record kind is + * discriminated by the id. Each document carries a {@code version} field used for optimistic + * concurrency. + * + *

The default {@link #getShardSize()} is {@value #DEFAULT_SHARD_SIZE} bytes, kept safely under + * MongoDB's 16 MB document-size limit; because the store forces a checkpoint whenever a diff + * would exceed the shard size, diff documents stay bounded too. + */ +public final class MongoSessionStoreOptions { + + /** Default database name. */ + public static final String DEFAULT_DATABASE = "genkit"; + + /** Default collection name. */ + public static final String DEFAULT_COLLECTION = "genkit_sessions"; + + /** Default number of turns between full checkpoints. */ + public static final int DEFAULT_CHECKPOINT_INTERVAL = 25; + + /** Default shard size in bytes for checkpoint state (1 MiB, under the 16 MB document cap). */ + public static final int DEFAULT_SHARD_SIZE = 1024 * 1024; + + /** Default subscription poll interval in milliseconds. */ + public static final long DEFAULT_POLL_INTERVAL_MS = 2000L; + + private final String databaseName; + private final String collectionName; + private final int checkpointInterval; + private final int shardSize; + private final Function snapshotPathPrefix; + private final long pollIntervalMs; + + private MongoSessionStoreOptions(Builder builder) { + this.databaseName = builder.databaseName; + this.collectionName = builder.collectionName; + this.checkpointInterval = builder.checkpointInterval; + this.shardSize = builder.shardSize; + this.snapshotPathPrefix = builder.snapshotPathPrefix; + this.pollIntervalMs = builder.pollIntervalMs; + } + + /** + * Returns the MongoDB database name (default {@value #DEFAULT_DATABASE}). + * + * @return the database name + */ + public String getDatabaseName() { + return databaseName; + } + + /** + * Returns the MongoDB collection name (default {@value #DEFAULT_COLLECTION}). + * + * @return the collection name + */ + public String getCollectionName() { + return collectionName; + } + + /** + * Returns the number of turns between full checkpoints (default {@value + * #DEFAULT_CHECKPOINT_INTERVAL}). + * + * @return the checkpoint interval + */ + public int getCheckpointInterval() { + return checkpointInterval; + } + + /** + * Returns the shard size in bytes for checkpoint state (default {@value #DEFAULT_SHARD_SIZE}). + * + * @return the shard size in bytes + */ + public int getShardSize() { + return shardSize; + } + + /** + * Returns the function that derives the per-tenant prefix from the per-request store options + * (default {@code o -> "global"}). + * + * @return the prefix function + */ + public Function getSnapshotPathPrefix() { + return snapshotPathPrefix; + } + + /** + * Returns the subscription poll interval in milliseconds (default {@value + * #DEFAULT_POLL_INTERVAL_MS}). + * + * @return the poll interval in milliseconds + */ + public long getPollIntervalMs() { + return pollIntervalMs; + } + + /** + * Returns default options. + * + * @return a {@code MongoSessionStoreOptions} with all defaults + */ + public static MongoSessionStoreOptions defaults() { + return builder().build(); + } + + /** + * Creates a new builder. + * + * @return a new builder + */ + public static Builder builder() { + return new Builder(); + } + + /** Builder for {@link MongoSessionStoreOptions}. */ + public static final class Builder { + private String databaseName = DEFAULT_DATABASE; + private String collectionName = DEFAULT_COLLECTION; + private int checkpointInterval = DEFAULT_CHECKPOINT_INTERVAL; + private int shardSize = DEFAULT_SHARD_SIZE; + private Function snapshotPathPrefix = o -> "global"; + private long pollIntervalMs = DEFAULT_POLL_INTERVAL_MS; + + private Builder() {} + + /** + * Sets the MongoDB database name. + * + * @param databaseName the database name + * @return this builder + */ + public Builder databaseName(String databaseName) { + this.databaseName = databaseName; + return this; + } + + /** + * Sets the MongoDB collection name. + * + * @param collectionName the collection name + * @return this builder + */ + public Builder collectionName(String collectionName) { + this.collectionName = collectionName; + return this; + } + + /** + * Sets the number of turns between full checkpoints. + * + * @param checkpointInterval the checkpoint interval (must be {@code >= 1}) + * @return this builder + */ + public Builder checkpointInterval(int checkpointInterval) { + this.checkpointInterval = checkpointInterval; + return this; + } + + /** + * Sets the shard size in bytes for checkpoint state (must stay under the 16 MB document cap). + * + * @param shardSize the shard size in bytes (must be {@code >= 1}) + * @return this builder + */ + public Builder shardSize(int shardSize) { + this.shardSize = shardSize; + return this; + } + + /** + * Sets the function that derives the per-tenant prefix. + * + * @param snapshotPathPrefix the prefix function + * @return this builder + */ + public Builder snapshotPathPrefix(Function snapshotPathPrefix) { + this.snapshotPathPrefix = snapshotPathPrefix; + return this; + } + + /** + * Sets the subscription poll interval in milliseconds. + * + * @param pollIntervalMs the poll interval (must be {@code >= 1}) + * @return this builder + */ + public Builder pollIntervalMs(long pollIntervalMs) { + this.pollIntervalMs = pollIntervalMs; + return this; + } + + /** + * Builds a new {@code MongoSessionStoreOptions}. + * + * @return a new options instance + */ + public MongoSessionStoreOptions build() { + if (databaseName == null || databaseName.isBlank()) { + throw new IllegalArgumentException("databaseName must be non-empty"); + } + if (collectionName == null || collectionName.isBlank()) { + throw new IllegalArgumentException("collectionName must be non-empty"); + } + if (checkpointInterval < 1) { + throw new IllegalArgumentException("checkpointInterval must be >= 1"); + } + if (shardSize < 1) { + throw new IllegalArgumentException("shardSize must be >= 1"); + } + if (snapshotPathPrefix == null) { + throw new IllegalArgumentException("snapshotPathPrefix must be non-null"); + } + if (pollIntervalMs < 1) { + throw new IllegalArgumentException("pollIntervalMs must be >= 1"); + } + return new MongoSessionStoreOptions(this); + } + } +} diff --git a/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/session/package-info.java b/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/session/package-info.java new file mode 100644 index 000000000..1c379b533 --- /dev/null +++ b/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/session/package-info.java @@ -0,0 +1,28 @@ +/* + * Copyright 2025 Google LLC + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * + * SPDX-License-Identifier: Apache-2.0 + */ + +/** + * MongoDB-backed agent session persistence. + * + *

{@link com.google.genkit.plugins.mongodb.session.MongoSessionStore} implements the Genkit + * {@code SessionStore} contract using the sharded checkpoint + RFC-6902 diff + pointer layout + * shared with the Firestore, DynamoDB, Cosmos DB, and PostgreSQL backends. Construct it directly + * from a {@code com.mongodb.client.MongoClient} and pass it to an agent via {@code + * AgentConfig.store(...)}. + */ +package com.google.genkit.plugins.mongodb.session; diff --git a/plugins/mongodb/src/test/java/com/google/genkit/plugins/mongodb/MongoPluginTest.java b/plugins/mongodb/src/test/java/com/google/genkit/plugins/mongodb/MongoPluginTest.java new file mode 100644 index 000000000..4ff62455d --- /dev/null +++ b/plugins/mongodb/src/test/java/com/google/genkit/plugins/mongodb/MongoPluginTest.java @@ -0,0 +1,65 @@ +/* + * Copyright 2025 Google LLC + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * + * SPDX-License-Identifier: Apache-2.0 + */ + +package com.google.genkit.plugins.mongodb; + +import static org.junit.jupiter.api.Assertions.assertEquals; +import static org.junit.jupiter.api.Assertions.assertThrows; + +import org.junit.jupiter.api.Test; + +/** Unit tests for {@link MongoPlugin}. */ +class MongoPluginTest { + + private static MongoVectorStoreConfig config() { + return MongoVectorStoreConfig.builder().collectionName("films").embedderName("e").build(); + } + + @Test + void getName() { + MongoPlugin plugin = + MongoPlugin.builder() + .connectionString("mongodb://localhost:27017") + .addCollection(config()) + .build(); + assertEquals("mongodb", plugin.getName()); + } + + @Test + void requiresConnectionStringOrClient() { + assertThrows( + IllegalStateException.class, () -> MongoPlugin.builder().addCollection(config()).build()); + } + + @Test + void requiresAtLeastOneCollection() { + assertThrows( + IllegalStateException.class, + () -> MongoPlugin.builder().connectionString("mongodb://localhost:27017").build()); + } + + @Test + void initWithoutRegistryThrows() { + MongoPlugin plugin = + MongoPlugin.builder() + .connectionString("mongodb://localhost:27017") + .addCollection(config()) + .build(); + assertThrows(IllegalStateException.class, plugin::init); + } +} diff --git a/plugins/mongodb/src/test/java/com/google/genkit/plugins/mongodb/MongoVectorStoreConfigTest.java b/plugins/mongodb/src/test/java/com/google/genkit/plugins/mongodb/MongoVectorStoreConfigTest.java new file mode 100644 index 000000000..2ee54c263 --- /dev/null +++ b/plugins/mongodb/src/test/java/com/google/genkit/plugins/mongodb/MongoVectorStoreConfigTest.java @@ -0,0 +1,94 @@ +/* + * Copyright 2025 Google LLC + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * + * SPDX-License-Identifier: Apache-2.0 + */ + +package com.google.genkit.plugins.mongodb; + +import static org.junit.jupiter.api.Assertions.assertEquals; +import static org.junit.jupiter.api.Assertions.assertThrows; +import static org.junit.jupiter.api.Assertions.assertTrue; + +import org.junit.jupiter.api.Test; + +/** Unit tests for {@link MongoVectorStoreConfig}. */ +class MongoVectorStoreConfigTest { + + @Test + void defaultsAreSane() { + MongoVectorStoreConfig c = + MongoVectorStoreConfig.builder() + .collectionName("films") + .embedderName("googleai/gemini-embedding-001") + .build(); + assertEquals("genkit", c.getDatabaseName()); + assertEquals("films", c.getCollectionName()); + assertEquals("genkit_vector_index", c.getIndexName()); + assertEquals(768, c.getDimension()); + assertEquals(MongoVectorStoreConfig.Similarity.COSINE, c.getSimilarity()); + assertEquals("cosine", c.getSimilarity().getValue()); + assertEquals("text", c.getTextField()); + assertEquals("embedding", c.getEmbeddingField()); + assertEquals(100, c.getNumCandidates()); + assertEquals(false, c.isCreateIndexIfNotExists()); + assertTrue(c.getAdditionalMetadata().isEmpty()); + } + + @Test + void customBuilder() { + MongoVectorStoreConfig c = + MongoVectorStoreConfig.builder() + .databaseName("rag") + .collectionName("docs") + .embedderName("e") + .indexName("idx") + .dimension(1536) + .similarity(MongoVectorStoreConfig.Similarity.DOT_PRODUCT) + .textField("content") + .embeddingField("vector") + .numCandidates(200) + .createIndexIfNotExists(true) + .addAdditionalMetadata("source", "wiki") + .build(); + assertEquals("rag", c.getDatabaseName()); + assertEquals("idx", c.getIndexName()); + assertEquals(1536, c.getDimension()); + assertEquals("dotProduct", c.getSimilarity().getValue()); + assertEquals("content", c.getTextField()); + assertEquals("vector", c.getEmbeddingField()); + assertEquals(200, c.getNumCandidates()); + assertEquals(true, c.isCreateIndexIfNotExists()); + assertEquals("wiki", c.getAdditionalMetadata().get("source")); + } + + @Test + void builderValidates() { + assertThrows( + IllegalArgumentException.class, + () -> MongoVectorStoreConfig.builder().embedderName("e").build()); + assertThrows( + IllegalArgumentException.class, + () -> MongoVectorStoreConfig.builder().collectionName("c").build()); + assertThrows( + IllegalArgumentException.class, + () -> + MongoVectorStoreConfig.builder() + .collectionName("c") + .embedderName("e") + .dimension(0) + .build()); + } +} diff --git a/plugins/mongodb/src/test/java/com/google/genkit/plugins/mongodb/MongoVectorStoreTest.java b/plugins/mongodb/src/test/java/com/google/genkit/plugins/mongodb/MongoVectorStoreTest.java new file mode 100644 index 000000000..0d2386fc0 --- /dev/null +++ b/plugins/mongodb/src/test/java/com/google/genkit/plugins/mongodb/MongoVectorStoreTest.java @@ -0,0 +1,138 @@ +/* + * Copyright 2025 Google LLC + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * + * SPDX-License-Identifier: Apache-2.0 + */ + +package com.google.genkit.plugins.mongodb; + +import static org.junit.jupiter.api.Assertions.assertEquals; +import static org.junit.jupiter.api.Assertions.assertFalse; +import static org.junit.jupiter.api.Assumptions.assumeTrue; + +import com.google.genkit.ai.Document; +import com.google.genkit.ai.EmbedResponse; +import com.google.genkit.ai.Embedder; +import com.google.genkit.ai.EmbedderInfo; +import com.google.genkit.ai.IndexerRequest; +import com.google.genkit.ai.RetrieverRequest; +import com.google.genkit.ai.RetrieverResponse; +import com.mongodb.client.MongoClient; +import com.mongodb.client.MongoClients; +import java.util.ArrayList; +import java.util.List; +import java.util.Random; +import java.util.UUID; +import org.junit.jupiter.api.AfterEach; +import org.junit.jupiter.api.BeforeEach; +import org.junit.jupiter.api.Test; + +/** + * Integration tests for {@link MongoVectorStore}, gated on the {@code MONGODB_ATLAS_URI} + * environment variable (a MongoDB Atlas cluster or the {@code mongodb/mongodb-atlas-local} Docker + * image). Skipped via {@link org.junit.jupiter.api.Assumptions} when unset. Uses a deterministic + * stub embedder so a query equal to an indexed document's text retrieves that document first. + */ +class MongoVectorStoreTest { + + private static final String URI = System.getenv("MONGODB_ATLAS_URI"); + private static final int DIM = 16; + + private MongoClient client; + private String database; + private String collection; + private MongoVectorStore store; + + private static boolean configured() { + return URI != null && !URI.isEmpty(); + } + + private static Embedder stubEmbedder() { + return new Embedder( + "test/stub", + new EmbedderInfo(), + (ctx, req) -> { + List out = new ArrayList<>(); + for (Document doc : req.getDocuments()) { + Random random = new Random(doc.text().hashCode()); + float[] values = new float[DIM]; + for (int i = 0; i < DIM; i++) { + values[i] = random.nextFloat(); + } + out.add(new EmbedResponse.Embedding(values)); + } + return new EmbedResponse(out); + }); + } + + @BeforeEach + void setUp() { + if (!configured()) { + return; + } + client = MongoClients.create(URI); + database = "genkit_test"; + collection = "vec_" + UUID.randomUUID().toString().replace("-", ""); + MongoVectorStoreConfig config = + MongoVectorStoreConfig.builder() + .databaseName(database) + .collectionName(collection) + .embedderName("test/stub") + .dimension(DIM) + .createIndexIfNotExists(true) + .build(); + store = new MongoVectorStore(client, config, stubEmbedder()); + } + + @AfterEach + void tearDown() { + if (client != null) { + if (database != null && collection != null) { + client.getDatabase(database).getCollection(collection).drop(); + } + client.close(); + } + } + + @Test + void indexThenRetrieveReturnsNearestFirst() throws Exception { + assumeTrue(configured()); + List docs = + List.of( + Document.fromText("The Matrix is a sci-fi film about simulated reality."), + Document.fromText("The Godfather is a crime film about a mafia family."), + Document.fromText("Inception is a sci-fi film about dreams.")); + store.index(null, new IndexerRequest(docs)); + + RetrieverRequest request = new RetrieverRequest(Document.fromText(docs.get(1).text())); + RetrieverRequest.RetrieverOptions options = new RetrieverRequest.RetrieverOptions(); + options.setK(3); + request.setOptions(options); + + // Newly indexed documents may take a moment to become searchable. + RetrieverResponse response = null; + for (int attempt = 0; attempt < 15; attempt++) { + response = store.retrieve(null, request); + if (!response.getDocuments().isEmpty()) { + break; + } + Thread.sleep(1000); + } + assertFalse(response.getDocuments().isEmpty()); + assertEquals( + "The Godfather is a crime film about a mafia family.", + response.getDocuments().get(0).text()); + } +} diff --git a/plugins/mongodb/src/test/java/com/google/genkit/plugins/mongodb/session/MongoSessionStoreOptionsTest.java b/plugins/mongodb/src/test/java/com/google/genkit/plugins/mongodb/session/MongoSessionStoreOptionsTest.java new file mode 100644 index 000000000..0dec9c964 --- /dev/null +++ b/plugins/mongodb/src/test/java/com/google/genkit/plugins/mongodb/session/MongoSessionStoreOptionsTest.java @@ -0,0 +1,78 @@ +/* + * Copyright 2025 Google LLC + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * + * SPDX-License-Identifier: Apache-2.0 + */ + +package com.google.genkit.plugins.mongodb.session; + +import static org.junit.jupiter.api.Assertions.assertEquals; +import static org.junit.jupiter.api.Assertions.assertThrows; + +import com.google.genkit.ai.agent.SessionStoreOptions; +import org.junit.jupiter.api.Test; + +/** Unit tests for {@link MongoSessionStoreOptions}. */ +class MongoSessionStoreOptionsTest { + + @Test + void defaultsAreSane() { + MongoSessionStoreOptions o = MongoSessionStoreOptions.defaults(); + assertEquals("genkit", o.getDatabaseName()); + assertEquals("genkit_sessions", o.getCollectionName()); + assertEquals(25, o.getCheckpointInterval()); + assertEquals(1024 * 1024, o.getShardSize()); + assertEquals("global", o.getSnapshotPathPrefix().apply(SessionStoreOptions.empty())); + assertEquals(2000L, o.getPollIntervalMs()); + } + + @Test + void customBuilder() { + MongoSessionStoreOptions o = + MongoSessionStoreOptions.builder() + .databaseName("my-db") + .collectionName("my_sessions") + .checkpointInterval(10) + .shardSize(4096) + .snapshotPathPrefix(so -> "tenant-1") + .pollIntervalMs(500) + .build(); + assertEquals("my-db", o.getDatabaseName()); + assertEquals("my_sessions", o.getCollectionName()); + assertEquals(10, o.getCheckpointInterval()); + assertEquals(4096, o.getShardSize()); + assertEquals("tenant-1", o.getSnapshotPathPrefix().apply(SessionStoreOptions.empty())); + assertEquals(500L, o.getPollIntervalMs()); + } + + @Test + void builderValidates() { + assertThrows( + IllegalArgumentException.class, + () -> MongoSessionStoreOptions.builder().databaseName("").build()); + assertThrows( + IllegalArgumentException.class, + () -> MongoSessionStoreOptions.builder().collectionName("").build()); + assertThrows( + IllegalArgumentException.class, + () -> MongoSessionStoreOptions.builder().checkpointInterval(0).build()); + assertThrows( + IllegalArgumentException.class, + () -> MongoSessionStoreOptions.builder().shardSize(0).build()); + assertThrows( + IllegalArgumentException.class, + () -> MongoSessionStoreOptions.builder().pollIntervalMs(0).build()); + } +} diff --git a/plugins/mongodb/src/test/java/com/google/genkit/plugins/mongodb/session/MongoSessionStoreTest.java b/plugins/mongodb/src/test/java/com/google/genkit/plugins/mongodb/session/MongoSessionStoreTest.java new file mode 100644 index 000000000..49d0d0f09 --- /dev/null +++ b/plugins/mongodb/src/test/java/com/google/genkit/plugins/mongodb/session/MongoSessionStoreTest.java @@ -0,0 +1,202 @@ +/* + * Copyright 2025 Google LLC + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * + * SPDX-License-Identifier: Apache-2.0 + */ + +package com.google.genkit.plugins.mongodb.session; + +import static org.junit.jupiter.api.Assertions.assertEquals; +import static org.junit.jupiter.api.Assertions.assertNotNull; +import static org.junit.jupiter.api.Assertions.assertNull; +import static org.junit.jupiter.api.Assertions.assertThrows; +import static org.junit.jupiter.api.Assumptions.assumeTrue; + +import com.google.genkit.ai.Message; +import com.google.genkit.ai.agent.GetSnapshotOptions; +import com.google.genkit.ai.agent.SessionSnapshot; +import com.google.genkit.ai.agent.SessionState; +import com.google.genkit.ai.agent.SessionStoreOptions; +import com.google.genkit.ai.agent.SnapshotStatus; +import com.google.genkit.core.GenkitException; +import com.mongodb.client.MongoClient; +import com.mongodb.client.MongoClients; +import java.util.HashMap; +import java.util.List; +import java.util.Map; +import java.util.UUID; +import org.junit.jupiter.api.AfterEach; +import org.junit.jupiter.api.BeforeEach; +import org.junit.jupiter.api.Test; + +/** + * Tests for {@link MongoSessionStore}. + * + *

Integration tests are gated on the {@code MONGO_URI} environment variable (e.g. a local + * MongoDB started with Docker). When it is unset the tests are skipped via {@link + * org.junit.jupiter.api.Assumptions}. + */ +class MongoSessionStoreTest { + + private static final String URI = System.getenv("MONGO_URI"); + + private MongoClient client; + private String database; + private String collection; + private MongoSessionStore> store; + + private static boolean configured() { + return URI != null && !URI.isEmpty(); + } + + @BeforeEach + void setUp() { + if (!configured()) { + return; // integration tests skip via assumeTrue + } + client = MongoClients.create(URI); + database = "genkit_test"; + collection = "genkit_sessions_test_" + UUID.randomUUID().toString().replace("-", ""); + store = + new MongoSessionStore<>( + client, + MongoSessionStoreOptions.builder() + .databaseName(database) + .collectionName(collection) + .checkpointInterval(3) + .build()); + } + + @AfterEach + void tearDown() { + if (client != null) { + if (database != null && collection != null) { + client.getDatabase(database).getCollection(collection).drop(); + } + client.close(); + } + } + + private static SessionSnapshot> snapshotWithState( + String sessionId, String parentId, Map custom) { + SessionState> state = + SessionState.>builder() + .sessionId(sessionId) + .messages(List.of(Message.user("hello"))) + .custom(custom) + .build(); + return SessionSnapshot.>builder() + .sessionId(sessionId) + .parentId(parentId) + .status(SnapshotStatus.COMPLETED) + .state(state) + .build(); + } + + @Test + void saveThenGetBySnapshotIdRoundTrips() { + assumeTrue(configured()); + String sessionId = "s-" + UUID.randomUUID(); + Map custom = new HashMap<>(); + custom.put("count", 1); + + String id = + store.saveSnapshot( + null, e -> snapshotWithState(sessionId, null, custom), SessionStoreOptions.empty()); + assertNotNull(id); + + SessionSnapshot> got = + store.getSnapshot(GetSnapshotOptions.builder().snapshotId(id).build()); + assertNotNull(got); + assertEquals(sessionId, got.getSessionId()); + assertEquals(SnapshotStatus.COMPLETED, got.getStatus()); + assertEquals(1, got.getState().getMessages().size()); + assertEquals(1, ((Number) got.getState().getCustom().get("count")).intValue()); + } + + @Test + void getBySessionIdReturnsLeaf() { + assumeTrue(configured()); + String sessionId = "s-" + UUID.randomUUID(); + + Map c1 = new HashMap<>(); + c1.put("count", 1); + String id1 = + store.saveSnapshot( + null, e -> snapshotWithState(sessionId, null, c1), SessionStoreOptions.empty()); + Map c2 = new HashMap<>(); + c2.put("count", 2); + String id2 = + store.saveSnapshot( + null, e -> snapshotWithState(sessionId, id1, c2), SessionStoreOptions.empty()); + + SessionSnapshot> latest = + store.getSnapshot(GetSnapshotOptions.builder().sessionId(sessionId).build()); + assertNotNull(latest); + assertEquals(id2, latest.getSnapshotId()); + assertEquals(2, ((Number) latest.getState().getCustom().get("count")).intValue()); + } + + @Test + void diffThenCheckpointReconstructs() { + assumeTrue(configured()); + // checkpointInterval is 3; save 5 turns across checkpoint boundaries and confirm the leaf + // reconstructs correctly (checkpoint shards + segment-path diffs). + String sessionId = "s-" + UUID.randomUUID(); + String parent = null; + String lastId = null; + for (int i = 1; i <= 5; i++) { + Map c = new HashMap<>(); + c.put("count", i); + final String p = parent; + lastId = + store.saveSnapshot( + null, e -> snapshotWithState(sessionId, p, c), SessionStoreOptions.empty()); + parent = lastId; + } + SessionSnapshot> got = + store.getSnapshot(GetSnapshotOptions.builder().snapshotId(lastId).build()); + assertNotNull(got); + assertEquals(5, ((Number) got.getState().getCustom().get("count")).intValue()); + } + + @Test + void rejectsEmptySessionId() { + assumeTrue(configured()); + GenkitException ex = + assertThrows( + GenkitException.class, + () -> + store.saveSnapshot( + null, + e -> snapshotWithState("", null, new HashMap<>()), + SessionStoreOptions.empty())); + assertEquals("INVALID_ARGUMENT", ex.getErrorCode()); + } + + @Test + void mutatorNullIsNoOp() { + assumeTrue(configured()); + assertNull(store.saveSnapshot(null, e -> null, SessionStoreOptions.empty())); + } + + @Test + void getUnknownSessionReturnsNull() { + assumeTrue(configured()); + assertNull( + store.getSnapshot( + GetSnapshotOptions.builder().sessionId("s-" + UUID.randomUUID()).build())); + } +} diff --git a/plugins/pinecone/README.md b/plugins/pinecone/README.md index d5c49172c..2300e4a9e 100644 --- a/plugins/pinecone/README.md +++ b/plugins/pinecone/README.md @@ -45,7 +45,7 @@ Genkit genkit = Genkit.builder() .apiKey(System.getenv("PINECONE_API_KEY")) .addIndex(PineconeIndexConfig.builder() .indexName("my-index") - .embedderName("googleai/text-embedding-004") + .embedderName("googleai/gemini-embedding-001") .dimension(768) .build()) .build()) @@ -133,12 +133,12 @@ PineconePlugin plugin = PineconePlugin.builder() .addIndex(PineconeIndexConfig.builder() .indexName("my-index") .namespace("production") - .embedderName("googleai/text-embedding-004") + .embedderName("googleai/gemini-embedding-001") .build()) .addIndex(PineconeIndexConfig.builder() .indexName("my-index") .namespace("staging") - .embedderName("googleai/text-embedding-004") + .embedderName("googleai/gemini-embedding-001") .build()) .build(); @@ -171,7 +171,7 @@ var ragFlow = genkit.defineFlow("ragFlow", String.class, String.class, (context, // Generate response using LLM GenerateRequest generateRequest = GenerateRequest.builder() - .model("googleai/gemini-2.0-flash") + .model("googleai/gemini-2.5-flash") .messages(List.of( Message.builder() .role(Role.USER) @@ -241,7 +241,7 @@ PineconeIndexConfig.Cloud.AZURE ```java PineconeIndexConfig.builder() .indexName("new-index") - .embedderName("googleai/text-embedding-004") + .embedderName("googleai/gemini-embedding-001") .dimension(768) .metric(PineconeIndexConfig.Metric.COSINE) .cloud(PineconeIndexConfig.Cloud.AWS) diff --git a/plugins/pinecone/src/main/java/com/google/genkit/plugins/pinecone/PineconeIndexConfig.java b/plugins/pinecone/src/main/java/com/google/genkit/plugins/pinecone/PineconeIndexConfig.java index 6ee72fb1c..f2561c08c 100644 --- a/plugins/pinecone/src/main/java/com/google/genkit/plugins/pinecone/PineconeIndexConfig.java +++ b/plugins/pinecone/src/main/java/com/google/genkit/plugins/pinecone/PineconeIndexConfig.java @@ -31,7 +31,7 @@ * 

{@code
  * PineconeIndexConfig config = PineconeIndexConfig.builder()
  *     .indexName("my-index")
- *     .embedderName("googleai/text-embedding-004")
+ *     .embedderName("googleai/gemini-embedding-001")
  *     .namespace("production")
  *     .build();
  * }
diff --git a/plugins/pinecone/src/main/java/com/google/genkit/plugins/pinecone/PineconePlugin.java b/plugins/pinecone/src/main/java/com/google/genkit/plugins/pinecone/PineconePlugin.java index f5b7c73ea..2848251db 100644 --- a/plugins/pinecone/src/main/java/com/google/genkit/plugins/pinecone/PineconePlugin.java +++ b/plugins/pinecone/src/main/java/com/google/genkit/plugins/pinecone/PineconePlugin.java @@ -45,7 +45,7 @@ * .addIndex( * PineconeIndexConfig.builder() * .indexName("my-index") - * .embedderName("googleai/text-embedding-004") + * .embedderName("googleai/gemini-embedding-001") * .build()) * .build()) * .build(); diff --git a/plugins/pinecone/src/main/java/com/google/genkit/plugins/pinecone/package-info.java b/plugins/pinecone/src/main/java/com/google/genkit/plugins/pinecone/package-info.java index 47a5f08c6..c591e6a8d 100644 --- a/plugins/pinecone/src/main/java/com/google/genkit/plugins/pinecone/package-info.java +++ b/plugins/pinecone/src/main/java/com/google/genkit/plugins/pinecone/package-info.java @@ -40,7 +40,7 @@ * .addIndex( * PineconeIndexConfig.builder() * .indexName("my-index") - * .embedderName("googleai/text-embedding-004") + * .embedderName("googleai/gemini-embedding-001") * .build()) * .build()) * .build(); diff --git a/plugins/postgresql/README.md b/plugins/postgresql/README.md index 5dca82e13..8304c9115 100644 --- a/plugins/postgresql/README.md +++ b/plugins/postgresql/README.md @@ -55,7 +55,7 @@ Genkit genkit = Genkit.builder() .password("pass") .addTable(PostgresTableConfig.builder() .tableName("documents") - .embedderName("googleai/text-embedding-004") + .embedderName("googleai/gemini-embedding-001") .vectorDimension(768) .build()) .build()) @@ -150,7 +150,7 @@ var ragFlow = genkit.defineFlow("ragFlow", String.class, String.class, (context, // Generate response using LLM GenerateRequest generateRequest = GenerateRequest.builder() - .model("googleai/gemini-2.0-flash") + .model("googleai/gemini-2.5-flash") .messages(List.of( Message.builder() .role(Role.USER) diff --git a/plugins/postgresql/src/main/java/com/google/genkit/plugins/postgresql/PostgresPlugin.java b/plugins/postgresql/src/main/java/com/google/genkit/plugins/postgresql/PostgresPlugin.java index cbac4d05e..78f22eea2 100644 --- a/plugins/postgresql/src/main/java/com/google/genkit/plugins/postgresql/PostgresPlugin.java +++ b/plugins/postgresql/src/main/java/com/google/genkit/plugins/postgresql/PostgresPlugin.java @@ -49,7 +49,7 @@ * .addTable( * PostgresTableConfig.builder() * .tableName("documents") - * .embedderName("googleai/text-embedding-004") + * .embedderName("googleai/gemini-embedding-001") * .build()) * .build()) * .build(); diff --git a/plugins/postgresql/src/main/java/com/google/genkit/plugins/postgresql/PostgresTableConfig.java b/plugins/postgresql/src/main/java/com/google/genkit/plugins/postgresql/PostgresTableConfig.java index 1df32c72a..8f22e1dab 100644 --- a/plugins/postgresql/src/main/java/com/google/genkit/plugins/postgresql/PostgresTableConfig.java +++ b/plugins/postgresql/src/main/java/com/google/genkit/plugins/postgresql/PostgresTableConfig.java @@ -31,7 +31,7 @@ * 
{@code
  * PostgresTableConfig config = PostgresTableConfig.builder()
  *     .tableName("documents")
- *     .embedderName("googleai/text-embedding-004")
+ *     .embedderName("googleai/gemini-embedding-001")
  *     .vectorDimension(768)
  *     .distanceStrategy(DistanceStrategy.COSINE)
  *     .build();
diff --git a/plugins/postgresql/src/main/java/com/google/genkit/plugins/postgresql/package-info.java b/plugins/postgresql/src/main/java/com/google/genkit/plugins/postgresql/package-info.java
index d29ff495c..f06ef3f7b 100644
--- a/plugins/postgresql/src/main/java/com/google/genkit/plugins/postgresql/package-info.java
+++ b/plugins/postgresql/src/main/java/com/google/genkit/plugins/postgresql/package-info.java
@@ -42,7 +42,7 @@
  *             .addTable(
  *                 PostgresTableConfig.builder()
  *                     .tableName("documents")
- *                     .embedderName("googleai/text-embedding-004")
+ *                     .embedderName("googleai/gemini-embedding-001")
  *                     .build())
  *             .build())
  *     .build();
diff --git a/plugins/postgresql/src/main/java/com/google/genkit/plugins/postgresql/session/PostgresSessionStore.java b/plugins/postgresql/src/main/java/com/google/genkit/plugins/postgresql/session/PostgresSessionStore.java
new file mode 100644
index 000000000..c7834503c
--- /dev/null
+++ b/plugins/postgresql/src/main/java/com/google/genkit/plugins/postgresql/session/PostgresSessionStore.java
@@ -0,0 +1,761 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package com.google.genkit.plugins.postgresql.session;
+
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.fasterxml.jackson.databind.node.ArrayNode;
+import com.fasterxml.jackson.databind.node.NullNode;
+import com.fasterxml.jackson.databind.node.ObjectNode;
+import com.google.genkit.ai.agent.AgentFinishReason;
+import com.google.genkit.ai.agent.GetSnapshotOptions;
+import com.google.genkit.ai.agent.RuntimeError;
+import com.google.genkit.ai.agent.SessionSnapshot;
+import com.google.genkit.ai.agent.SessionState;
+import com.google.genkit.ai.agent.SessionStore;
+import com.google.genkit.ai.agent.SessionStoreOptions;
+import com.google.genkit.ai.agent.SnapshotMutator;
+import com.google.genkit.ai.agent.SnapshotStatus;
+import com.google.genkit.ai.agent.SnapshotSubscriber;
+import com.google.genkit.ai.agent.internal.SnapshotSharding;
+import com.google.genkit.core.GenkitException;
+import com.google.genkit.core.JsonUtils;
+import com.google.genkit.core.jsonpatch.JsonPatch;
+import java.nio.charset.StandardCharsets;
+import java.sql.Connection;
+import java.sql.PreparedStatement;
+import java.sql.ResultSet;
+import java.sql.SQLException;
+import java.sql.Statement;
+import java.util.ArrayList;
+import java.util.List;
+import java.util.UUID;
+import java.util.concurrent.Executors;
+import java.util.concurrent.ScheduledExecutorService;
+import java.util.concurrent.ScheduledFuture;
+import java.util.concurrent.TimeUnit;
+import java.util.function.Consumer;
+import javax.sql.DataSource;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+/**
+ * PostgreSQL-backed implementation of {@link SessionStore} and {@link SnapshotSubscriber}.
+ *
+ * 
Persists session snapshots with the same sharded checkpoint + diff + pointer layout as the
+ * Firestore, DynamoDB, and Cosmos DB backends (see {@code FirestoreSessionStore}), sharing the pure
+ * logic in {@link SnapshotSharding}.
+ *
+ * 
Storage layout (single table)

+ *
+ * 
All records live in one table (default {@code genkit_sessions}) keyed by {@code (pk, id)}. The
+ * {@code pk} column holds the per-tenant prefix (default {@code "global"}); the {@code id} column
+ * discriminates the record kind, each row carrying a JSONB {@code doc} payload and a {@code
+ * version} counter:
+ *
+ * 
    
+ *   
  • {@code SNAP_} — one metadata record per snapshot. {@code kind} is {@code
+ *       "checkpoint"} or {@code "diff"}; carries {@code checkpointId}, {@code
+ *       checkpointShardCount}, {@code segmentPath}, {@code statePatch} (RFC-6902 as a JSON string
+ *       for diffs) and {@code error} (JSON string).
+ *   
  • {@code SHARD__} — a shard of the checkpoint state JSON.
+ *   
  • {@code PTR_} — the current leaf pointer for a session.
+ * 

+ *
+ * 
Concurrency

+ *
+ * 
Shard and snapshot records are idempotent by key (upserted); updating an existing snapshot id
+ * uses a {@code version} conditional update, re-applying the (pure) mutator on conflict. The
+ * session pointer is advanced monotonically (never backward) under the same {@code version}
+ * concurrency. Shards are written before the snapshot record, which is written before the pointer
+ * flips, so a reader following the pointer always sees complete data.
+ *
+ * @param  the type of custom session state
+ */
+public final class PostgresSessionStore implements SessionStore, SnapshotSubscriber {
+
+  private static final Logger logger = LoggerFactory.getLogger(PostgresSessionStore.class);
+  private static final ObjectMapper MAPPER = JsonUtils.getObjectMapper();
+
+  static final String KIND_CHECKPOINT = "checkpoint";
+  static final String KIND_DIFF = "diff";
+
+  private static final int MAX_ATTEMPTS = 5;
+
+  private final DataSource dataSource;
+  private final PostgresSessionStoreOptions options;
+  private final String table;
+  private final ScheduledExecutorService scheduler;
+
+  /**
+   * Creates a store with default options.
+   *
+   * @param dataSource the PostgreSQL data source
+   */
+  public PostgresSessionStore(DataSource dataSource) {
+    this(dataSource, PostgresSessionStoreOptions.defaults());
+  }
+
+  /**
+   * Creates a store.
+   *
+   * @param dataSource the PostgreSQL data source
+   * @param options the store options
+   */
+  public PostgresSessionStore(DataSource dataSource, PostgresSessionStoreOptions options) {
+    if (dataSource == null) {
+      throw new IllegalArgumentException("DataSource must be non-null");
+    }
+    if (options == null) {
+      throw new IllegalArgumentException("options must be non-null");
+    }
+    this.dataSource = dataSource;
+    this.options = options;
+    this.table = quoteIdentifier(options.getTableName());
+    if (options.isCreateTableIfNotExists()) {
+      ensureTable();
+    }
+    this.scheduler =
+        Executors.newSingleThreadScheduledExecutor(
+            r -> {
+              Thread t = new Thread(r, "genkit-postgres-session-store-poll");
+              t.setDaemon(true);
+              return t;
+            });
+  }
+
+  private void ensureTable() {
+    String sql =
+        "CREATE TABLE IF NOT EXISTS "
+            + table
+            + " ("
+            + "pk TEXT NOT NULL,"
+            + "id TEXT NOT NULL,"
+            + "doc JSONB NOT NULL,"
+            + "version BIGINT NOT NULL DEFAULT 1,"
+            + "PRIMARY KEY (pk, id))";
+    try (Connection conn = dataSource.getConnection();
+        Statement stmt = conn.createStatement()) {
+      stmt.execute(sql);
+      logger.info("PostgreSQL session store initialized for table: {}", options.getTableName());
+    } catch (SQLException e) {
+      throw new GenkitException("Failed to create session table: " + e.getMessage(), e);
+    }
+  }
+
+  // ──────────────────────────────────────────────────────────────────────────
+  // Id helpers
+  // ──────────────────────────────────────────────────────────────────────────
+
+  private String prefix(SessionStoreOptions opts) {
+    String p =
+        options.getSnapshotPathPrefix().apply(opts != null ? opts : SessionStoreOptions.empty());
+    return (p == null || p.isBlank()) ? "global" : p;
+  }
+
+  private static String snapId(String snapshotId) {
+    return "SNAP_" + snapshotId;
+  }
+
+  private static String shardId(String checkpointId, int index) {
+    return "SHARD_" + checkpointId + "_" + index;
+  }
+
+  private static String ptrId(String sessionId) {
+    return "PTR_" + sessionId;
+  }
+
+  // ──────────────────────────────────────────────────────────────────────────
+  // SnapshotWriter
+  // ──────────────────────────────────────────────────────────────────────────
+
+  /**
+   * {@inheritDoc}
+   *
+   * 
Implements the same identity/sessionId/status defaulting contract as the reference stores,
+   * then writes the snapshot (checkpoint shards + metadata record) and advances the session
+   * pointer.
+   */
+  @Override
+  public String saveSnapshot(
+      String snapshotId, SnapshotMutator mutator, SessionStoreOptions storeOpts) {
+    String prefix = prefix(storeOpts);
+
+    for (int attempt = 1; ; attempt++) {
+      // 1. Read existing snapshot (+ version) and reconstruct state.
+      SessionSnapshot existing = null;
+      String existingSessionId = null;
+      Long existingVersion = null;
+      if (snapshotId != null) {
+        SnapshotSharding.validateId(snapshotId);
+        Row row = readRow(prefix, snapId(snapshotId));
+        if (row != null) {
+          existing = readSnapshot(prefix, row.doc);
+          existingSessionId = existing.getSessionId();
+          existingVersion = row.version;
+        }
+      }
+
+      // 2. Apply mutator (pure — safe to re-run on conflict retry).
+      SessionSnapshot result = mutator.apply(existing);
+      if (result == null) {
+        return null;
+      }
+
+      // 3. Identity / sessionId / status defaulting (mirror the reference stores).
+      String finalId;
+      if (snapshotId != null) {
+        finalId = snapshotId;
+      } else if (result.getSnapshotId() != null && !result.getSnapshotId().isBlank()) {
+        finalId = result.getSnapshotId();
+      } else {
+        finalId = UUID.randomUUID().toString();
+      }
+      SnapshotSharding.validateId(finalId);
+      result.setSnapshotId(finalId);
+
+      if (existingSessionId != null) {
+        result.setSessionId(existingSessionId);
+      }
+      if (result.getSessionId() == null && result.getState() != null) {
+        result.setSessionId(result.getState().getSessionId());
+      }
+      if (result.getSessionId() == null || result.getSessionId().isBlank()) {
+        throw GenkitException.builder()
+            .message("snapshot requires sessionId")
+            .errorCode("INVALID_ARGUMENT")
+            .build();
+      }
+      if (result.getStatus() == null) {
+        result.setStatus(SnapshotStatus.COMPLETED);
+      }
+
+      // 4. Resolve parent metadata to decide checkpoint vs diff.
+      String parentId = result.getParentId();
+      JsonNode newState = stateToJson(result.getState());
+
+      ParentInfo parent = null;
+      if (parentId != null && !parentId.isBlank()) {
+        Row parentRow = readRow(prefix, snapId(parentId));
+        if (parentRow != null) {
+          parent = loadParentInfo(prefix, parentRow.doc);
+        }
+      }
+
+      boolean parentExists = parent != null;
+      int depthFromCheckpoint = 0;
+      JsonNode statePatch = null;
+      int diffSizeBytes = 0;
+      if (parent != null) {
+        depthFromCheckpoint = parent.segmentPath.size() + 1;
+        statePatch = JsonPatch.diff(parent.state, newState);
+        diffSizeBytes = jsonBytes(statePatch);
+      }
+
+      boolean checkpoint =
+          SnapshotSharding.shouldCheckpoint(
+              parentExists,
+              depthFromCheckpoint,
+              options.getCheckpointInterval(),
+              diffSizeBytes,
+              options.getShardSize());
+
+      // 5. Build the snapshot metadata record (+ shard rows for a checkpoint).
+      ObjectNode doc = baseDoc(result);
+      String checkpointId;
+      int checkpointShardCount;
+      List segmentPath;
+
+      if (checkpoint) {
+        checkpointId = finalId;
+        String stateJson = writeJson(newState);
+        List shards = SnapshotSharding.shardString(stateJson, options.getShardSize());
+        checkpointShardCount = shards.size();
+        segmentPath = new ArrayList<>();
+        for (int i = 0; i < shards.size(); i++) {
+          ObjectNode shardDoc = MAPPER.createObjectNode();
+          shardDoc.put("checkpointId", checkpointId);
+          shardDoc.put("index", i);
+          shardDoc.put("data", shards.get(i));
+          upsertRow(prefix, shardId(checkpointId, i), shardDoc);
+        }
+        doc.put("kind", KIND_CHECKPOINT);
+      } else {
+        ParentInfo p = parent;
+        if (p == null) {
+          throw new GenkitException("internal: diff path without parent");
+        }
+        checkpointId = p.checkpointId;
+        checkpointShardCount = p.checkpointShardCount;
+        segmentPath = new ArrayList<>(p.segmentPath);
+        segmentPath.add(finalId);
+        doc.put("kind", KIND_DIFF);
+        doc.put("statePatch", writeJson(statePatch));
+      }
+
+      doc.put("checkpointId", checkpointId);
+      doc.put("checkpointShardCount", checkpointShardCount);
+      putStringArray(doc, "segmentPath", segmentPath);
+
+      // 6. Write the snapshot record with optimistic concurrency.
+      boolean written = writeConditional(prefix, snapId(finalId), doc, existingVersion);
+      if (!written) {
+        if (attempt < MAX_ATTEMPTS) {
+          continue; // concurrent writer won the race; re-read and retry the pure mutator.
+        }
+        throw new GenkitException("Failed to save snapshot after " + MAX_ATTEMPTS + " attempts");
+      }
+
+      // 7. Advance the session pointer (never backward).
+      advancePointer(
+          prefix,
+          result.getSessionId(),
+          finalId,
+          result.getCreatedAt(),
+          result.getUpdatedAt(),
+          checkpointId,
+          checkpointShardCount,
+          segmentPath);
+
+      return finalId;
+    }
+  }
+
+  /** Advances the session pointer to the new leaf unless the stored pointer is already newer. */
+  private void advancePointer(
+      String prefix,
+      String sessionId,
+      String snapshotId,
+      String createdAt,
+      String updatedAt,
+      String checkpointId,
+      int checkpointShardCount,
+      List segmentPath) {
+    String id = ptrId(sessionId);
+
+    for (int attempt = 1; ; attempt++) {
+      Row existing = readRow(prefix, id);
+      Long version = existing != null ? existing.version : null;
+      if (existing != null) {
+        String currentLeaf = getString(existing.doc, "currentSnapshotId");
+        String currentCreatedAt = getString(existing.doc, "currentCreatedAt");
+        boolean sameLeaf = snapshotId.equals(currentLeaf);
+        boolean newer =
+            createdAt == null
+                || currentCreatedAt == null
+                || createdAt.compareTo(currentCreatedAt) >= 0;
+        if (!sameLeaf && !newer) {
+          return; // stored pointer is already newer; don't move backward.
+        }
+      }
+
+      ObjectNode ptr = MAPPER.createObjectNode();
+      ptr.put("currentSnapshotId", snapshotId);
+      ptr.put("checkpointId", checkpointId);
+      ptr.put("checkpointShardCount", checkpointShardCount);
+      putStringArray(ptr, "segmentPath", segmentPath);
+      if (createdAt != null) {
+        ptr.put("currentCreatedAt", createdAt);
+      }
+      if (updatedAt != null) {
+        ptr.put("updatedAt", updatedAt);
+      }
+
+      if (writeConditional(prefix, id, ptr, version)) {
+        return;
+      }
+      if (attempt >= MAX_ATTEMPTS) {
+        logger.debug("Pointer for session {} not advanced (lost concurrency race)", sessionId);
+        return;
+      }
+    }
+  }
+
+  // ──────────────────────────────────────────────────────────────────────────
+  // SnapshotReader
+  // ──────────────────────────────────────────────────────────────────────────
+
+  /**
+   * {@inheritDoc}
+   *
+   * 
By {@code snapshotId}: loads the snapshot record and reconstructs its state from the
+   * checkpoint shards + ordered {@code segmentPath} diffs. By {@code sessionId}: reads the pointer,
+   * then loads and reconstructs the pointed snapshot.
+   */
+  @Override
+  public SessionSnapshot getSnapshot(GetSnapshotOptions opts) {
+    if (opts == null) {
+      return null;
+    }
+    String prefix = prefix(SessionStoreOptions.empty());
+
+    if (opts.getSnapshotId() != null) {
+      SnapshotSharding.validateId(opts.getSnapshotId());
+      Row row = readRow(prefix, snapId(opts.getSnapshotId()));
+      return row == null ? null : readSnapshot(prefix, row.doc);
+    }
+    if (opts.getSessionId() != null) {
+      SnapshotSharding.validateId(opts.getSessionId());
+      Row pointer = readRow(prefix, ptrId(opts.getSessionId()));
+      if (pointer == null) {
+        return null;
+      }
+      String leafId = getString(pointer.doc, "currentSnapshotId");
+      if (leafId == null) {
+        return null;
+      }
+      Row row = readRow(prefix, snapId(leafId));
+      return row == null ? null : readSnapshot(prefix, row.doc);
+    }
+    return null;
+  }
+
+  // ──────────────────────────────────────────────────────────────────────────
+  // SnapshotSubscriber (polling)
+  // ──────────────────────────────────────────────────────────────────────────
+
+  /**
+   * {@inheritDoc}
+   *
+   * 
The subscription polls {@link #getSnapshot} on a shared daemon scheduler and fires the
+   * callback whenever the serialized snapshot content changes. The callback also fires immediately
+   * if the snapshot already exists.
+   */
+  @Override
+  public AutoCloseable onSnapshotStateChange(
+      String snapshotId, Consumer> cb, SessionStoreOptions storeOpts) {
+    SnapshotSharding.validateId(snapshotId);
+    GetSnapshotOptions get = GetSnapshotOptions.builder().snapshotId(snapshotId).build();
+
+    final String[] lastContent = {null};
+    SessionSnapshot initial = getSnapshot(get);
+    if (initial != null) {
+      lastContent[0] = serializeQuietly(initial);
+      cb.accept(initial);
+    }
+
+    ScheduledFuture future =
+        scheduler.scheduleAtFixedRate(
+            () -> {
+              try {
+                SessionSnapshot snap = getSnapshot(get);
+                if (snap == null) {
+                  return;
+                }
+                String content = serializeQuietly(snap);
+                if (!content.equals(lastContent[0])) {
+                  lastContent[0] = content;
+                  cb.accept(snap);
+                }
+              } catch (Exception e) {
+                // Swallow poll errors — don't kill the scheduler thread.
+              }
+            },
+            options.getPollIntervalMs(),
+            options.getPollIntervalMs(),
+            TimeUnit.MILLISECONDS);
+
+    return () -> future.cancel(false);
+  }
+
+  // ──────────────────────────────────────────────────────────────────────────
+  // Document (de)serialization
+  // ──────────────────────────────────────────────────────────────────────────
+
+  /** Builds the non-state base document for a snapshot (metadata only). */
+  private ObjectNode baseDoc(SessionSnapshot snap) {
+    ObjectNode data = MAPPER.createObjectNode();
+    data.put("snapshotId", snap.getSnapshotId());
+    data.put("sessionId", snap.getSessionId());
+    if (snap.getParentId() != null) {
+      data.put("parentId", snap.getParentId());
+    }
+    if (snap.getCreatedAt() != null) {
+      data.put("createdAt", snap.getCreatedAt());
+    }
+    if (snap.getUpdatedAt() != null) {
+      data.put("updatedAt", snap.getUpdatedAt());
+    }
+    if (snap.getHeartbeatAt() != null) {
+      data.put("heartbeatAt", snap.getHeartbeatAt());
+    }
+    if (snap.getStatus() != null) {
+      data.put("status", snap.getStatus().getValue());
+    }
+    if (snap.getFinishReason() != null) {
+      data.put("finishReason", snap.getFinishReason().getValue());
+    }
+    if (snap.getError() != null) {
+      data.put("error", writeJson(snap.getError()));
+    }
+    return data;
+  }
+
+  /** Holds the reconstructed parent state and its checkpoint lineage. */
+  private static final class ParentInfo {
+    JsonNode state;
+    String checkpointId;
+    int checkpointShardCount;
+    List segmentPath;
+  }
+
+  /** Loads a parent snapshot's checkpoint lineage and reconstructs its state. */
+  private ParentInfo loadParentInfo(String prefix, ObjectNode doc) {
+    ParentInfo info = new ParentInfo();
+    info.checkpointId = getString(doc, "checkpointId");
+    info.checkpointShardCount = getInt(doc, "checkpointShardCount");
+    info.segmentPath = getStringList(doc, "segmentPath");
+    info.state =
+        reconstructFullState(
+            prefix, info.checkpointId, info.checkpointShardCount, info.segmentPath);
+    return info;
+  }
+
+  /** Reads and fully reconstructs a snapshot from its metadata document. */
+  private SessionSnapshot readSnapshot(String prefix, ObjectNode doc) {
+    String checkpointId = getString(doc, "checkpointId");
+    int checkpointShardCount = getInt(doc, "checkpointShardCount");
+    List segmentPath = getStringList(doc, "segmentPath");
+    JsonNode state = reconstructFullState(prefix, checkpointId, checkpointShardCount, segmentPath);
+    return docToSnapshot(doc, state);
+  }
+
+  /**
+   * Reconstructs full state: loads the checkpoint shards (concatenate, parse) then applies the
+   * {@code segmentPath} diffs in order.
+   */
+  private JsonNode reconstructFullState(
+      String prefix, String checkpointId, int checkpointShardCount, List segmentPath) {
+    if (checkpointId == null) {
+      return NullNode.getInstance();
+    }
+    List shardContents = new ArrayList<>();
+    for (int i = 0; i < checkpointShardCount; i++) {
+      Row shard = readRow(prefix, shardId(checkpointId, i));
+      shardContents.add(shard != null ? getString(shard.doc, "data") : "");
+    }
+    String checkpointJson = SnapshotSharding.reassembleShards(shardContents);
+
+    List diffs = new ArrayList<>();
+    for (String diffId : segmentPath) {
+      Row diffRow = readRow(prefix, snapId(diffId));
+      if (diffRow != null) {
+        String patch = getString(diffRow.doc, "statePatch");
+        if (patch != null) {
+          diffs.add(patch);
+        }
+      }
+    }
+    try {
+      return SnapshotSharding.reconstructState(checkpointJson, diffs);
+    } catch (Exception e) {
+      throw new GenkitException("Failed to reconstruct session state: " + e.getMessage(), e);
+    }
+  }
+
+  /** Builds a {@link SessionSnapshot} from a metadata document and reconstructed state. */
+  @SuppressWarnings("unchecked")
+  private SessionSnapshot docToSnapshot(ObjectNode doc, JsonNode state) {
+    SessionSnapshot.Builder builder = SessionSnapshot.builder();
+    builder.snapshotId(getString(doc, "snapshotId"));
+    builder.sessionId(getString(doc, "sessionId"));
+    builder.parentId(getString(doc, "parentId"));
+    builder.createdAt(getString(doc, "createdAt"));
+    builder.updatedAt(getString(doc, "updatedAt"));
+    builder.heartbeatAt(getString(doc, "heartbeatAt"));
+    String status = getString(doc, "status");
+    if (status != null) {
+      builder.status(SnapshotStatus.fromValueOrCompleted(status));
+    }
+    String finishReason = getString(doc, "finishReason");
+    if (finishReason != null) {
+      builder.finishReason(AgentFinishReason.fromValue(finishReason));
+    }
+    String errorJson = getString(doc, "error");
+    if (errorJson != null) {
+      try {
+        builder.error(MAPPER.readValue(errorJson, RuntimeError.class));
+      } catch (Exception e) {
+        throw new GenkitException("Failed to parse snapshot error: " + e.getMessage(), e);
+      }
+    }
+    if (state != null && !state.isNull()) {
+      try {
+        builder.state((SessionState) MAPPER.treeToValue(state, SessionState.class));
+      } catch (Exception e) {
+        throw new GenkitException("Failed to parse session state: " + e.getMessage(), e);
+      }
+    }
+    return builder.build();
+  }
+
+  /** Serializes session state to a JSON node (null state → JSON null). */
+  private JsonNode stateToJson(SessionState state) {
+    if (state == null) {
+      return NullNode.getInstance();
+    }
+    return MAPPER.valueToTree(state);
+  }
+
+  // ──────────────────────────────────────────────────────────────────────────
+  // Low-level JDBC + node helpers
+  // ──────────────────────────────────────────────────────────────────────────
+
+  /** A single stored row: its JSON document and optimistic-concurrency version. */
+  private static final class Row {
+    final ObjectNode doc;
+    final long version;
+
+    Row(ObjectNode doc, long version) {
+      this.doc = doc;
+      this.version = version;
+    }
+  }
+
+  /** Reads a row by key, or {@code null} when it does not exist. */
+  private Row readRow(String prefix, String id) {
+    String sql = "SELECT doc, version FROM " + table + " WHERE pk = ? AND id = ?";
+    try (Connection conn = dataSource.getConnection();
+        PreparedStatement pstmt = conn.prepareStatement(sql)) {
+      pstmt.setString(1, prefix);
+      pstmt.setString(2, id);
+      try (ResultSet rs = pstmt.executeQuery()) {
+        if (!rs.next()) {
+          return null;
+        }
+        String docJson = rs.getString(1);
+        long version = rs.getLong(2);
+        return new Row((ObjectNode) MAPPER.readTree(docJson), version);
+      }
+    } catch (Exception e) {
+      throw new GenkitException("Failed to read session row " + id + ": " + e.getMessage(), e);
+    }
+  }
+
+  /** Idempotently upserts a row (used for shard records), bumping the version. */
+  private void upsertRow(String prefix, String id, ObjectNode doc) {
+    String sql =
+        "INSERT INTO "
+            + table
+            + " (pk, id, doc, version) VALUES (?, ?, ?::jsonb, 1) "
+            + "ON CONFLICT (pk, id) DO UPDATE SET doc = EXCLUDED.doc, version = "
+            + table
+            + ".version + 1";
+    try (Connection conn = dataSource.getConnection();
+        PreparedStatement pstmt = conn.prepareStatement(sql)) {
+      pstmt.setString(1, prefix);
+      pstmt.setString(2, id);
+      pstmt.setString(3, writeJson(doc));
+      pstmt.executeUpdate();
+    } catch (SQLException e) {
+      throw new GenkitException("Failed to upsert session row " + id + ": " + e.getMessage(), e);
+    }
+  }
+
+  /**
+   * Writes a row with optimistic concurrency: inserts when {@code expectedVersion} is {@code null},
+   * otherwise updates only when the stored version still matches. Returns {@code false} when a
+   * concurrent writer won the race (the caller re-reads and retries the pure mutator).
+   */
+  private boolean writeConditional(String prefix, String id, ObjectNode doc, Long expectedVersion) {
+    String docJson = writeJson(doc);
+    if (expectedVersion == null) {
+      String sql =
+          "INSERT INTO "
+              + table
+              + " (pk, id, doc, version) VALUES (?, ?, ?::jsonb, 1) ON CONFLICT (pk, id) DO NOTHING";
+      try (Connection conn = dataSource.getConnection();
+          PreparedStatement pstmt = conn.prepareStatement(sql)) {
+        pstmt.setString(1, prefix);
+        pstmt.setString(2, id);
+        pstmt.setString(3, docJson);
+        return pstmt.executeUpdate() == 1;
+      } catch (SQLException e) {
+        throw new GenkitException("Failed to insert session row " + id + ": " + e.getMessage(), e);
+      }
+    }
+    String sql =
+        "UPDATE "
+            + table
+            + " SET doc = ?::jsonb, version = version + 1 WHERE pk = ? AND id = ? AND version = ?";
+    try (Connection conn = dataSource.getConnection();
+        PreparedStatement pstmt = conn.prepareStatement(sql)) {
+      pstmt.setString(1, docJson);
+      pstmt.setString(2, prefix);
+      pstmt.setString(3, id);
+      pstmt.setLong(4, expectedVersion);
+      return pstmt.executeUpdate() == 1;
+    } catch (SQLException e) {
+      throw new GenkitException("Failed to update session row " + id + ": " + e.getMessage(), e);
+    }
+  }
+
+  private static void putStringArray(ObjectNode node, String name, List values) {
+    ArrayNode arr = node.putArray(name);
+    for (String v : values) {
+      arr.add(v);
+    }
+  }
+
+  private static String getString(ObjectNode node, String name) {
+    JsonNode v = node.get(name);
+    return (v == null || v.isNull()) ? null : v.asText();
+  }
+
+  private static int getInt(ObjectNode node, String name) {
+    JsonNode v = node.get(name);
+    return (v == null || v.isNull()) ? 0 : v.asInt();
+  }
+
+  private static List getStringList(ObjectNode node, String name) {
+    List out = new ArrayList<>();
+    JsonNode v = node.get(name);
+    if (v != null && v.isArray()) {
+      for (JsonNode e : v) {
+        out.add(e.asText());
+      }
+    }
+    return out;
+  }
+
+  private static String writeJson(Object value) {
+    try {
+      return MAPPER.writeValueAsString(value);
+    } catch (Exception e) {
+      throw new GenkitException("Failed to serialize value: " + e.getMessage(), e);
+    }
+  }
+
+  private static int jsonBytes(Object value) {
+    return writeJson(value).getBytes(StandardCharsets.UTF_8).length;
+  }
+
+  private static String serializeQuietly(SessionSnapshot snap) {
+    try {
+      return MAPPER.writeValueAsString(MAPPER.valueToTree(snap));
+    } catch (Exception e) {
+      return "";
+    }
+  }
+
+  /** Quotes a SQL identifier, doubling embedded double quotes. */
+  private static String quoteIdentifier(String identifier) {
+    return "\"" + identifier.replace("\"", "\"\"") + "\"";
+  }
+}
diff --git a/plugins/postgresql/src/main/java/com/google/genkit/plugins/postgresql/session/PostgresSessionStoreOptions.java b/plugins/postgresql/src/main/java/com/google/genkit/plugins/postgresql/session/PostgresSessionStoreOptions.java
new file mode 100644
index 000000000..c79cafaa5
--- /dev/null
+++ b/plugins/postgresql/src/main/java/com/google/genkit/plugins/postgresql/session/PostgresSessionStoreOptions.java
@@ -0,0 +1,243 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package com.google.genkit.plugins.postgresql.session;
+
+import com.google.genkit.ai.agent.SessionStoreOptions;
+import java.util.function.Function;
+
+/**
+ * Configuration for {@link PostgresSessionStore}.
+ *
+ * 
The store persists all records in a single PostgreSQL table (default {@value #DEFAULT_TABLE})
+ * keyed by {@code (pk, id)}. The {@code pk} column holds the per-tenant prefix (default {@code
+ * "global"}); the {@code id} column discriminates snapshot, shard, and pointer records. Each row
+ * carries a JSONB {@code doc} payload and a monotonically increasing {@code version} used for
+ * optimistic concurrency.
+ *
+ * 
The default {@link #getShardSize()} is {@value #DEFAULT_SHARD_SIZE} bytes; because the store
+ * forces a checkpoint whenever a diff would exceed the shard size, diff rows stay bounded too.
+ */
+public final class PostgresSessionStoreOptions {
+
+  /** Default table name. */
+  public static final String DEFAULT_TABLE = "genkit_sessions";
+
+  /** Default number of turns between full checkpoints. */
+  public static final int DEFAULT_CHECKPOINT_INTERVAL = 25;
+
+  /** Default shard size in bytes for checkpoint state (1 MiB). */
+  public static final int DEFAULT_SHARD_SIZE = 1024 * 1024;
+
+  /** Default subscription poll interval in milliseconds. */
+  public static final long DEFAULT_POLL_INTERVAL_MS = 2000L;
+
+  private final String tableName;
+  private final int checkpointInterval;
+  private final int shardSize;
+  private final Function snapshotPathPrefix;
+  private final boolean createTableIfNotExists;
+  private final long pollIntervalMs;
+
+  private PostgresSessionStoreOptions(Builder builder) {
+    this.tableName = builder.tableName;
+    this.checkpointInterval = builder.checkpointInterval;
+    this.shardSize = builder.shardSize;
+    this.snapshotPathPrefix = builder.snapshotPathPrefix;
+    this.createTableIfNotExists = builder.createTableIfNotExists;
+    this.pollIntervalMs = builder.pollIntervalMs;
+  }
+
+  /**
+   * Returns the PostgreSQL table name (default {@value #DEFAULT_TABLE}).
+   *
+   * @return the table name
+   */
+  public String getTableName() {
+    return tableName;
+  }
+
+  /**
+   * Returns the number of turns between full checkpoints (default {@value
+   * #DEFAULT_CHECKPOINT_INTERVAL}).
+   *
+   * @return the checkpoint interval
+   */
+  public int getCheckpointInterval() {
+    return checkpointInterval;
+  }
+
+  /**
+   * Returns the shard size in bytes for checkpoint state (default {@value #DEFAULT_SHARD_SIZE}).
+   *
+   * @return the shard size in bytes
+   */
+  public int getShardSize() {
+    return shardSize;
+  }
+
+  /**
+   * Returns the function that derives the per-tenant prefix from the per-request store options
+   * (default {@code o -> "global"}).
+   *
+   * @return the prefix function
+   */
+  public Function getSnapshotPathPrefix() {
+    return snapshotPathPrefix;
+  }
+
+  /**
+   * Returns whether the store should create the table on first use if it does not exist (default
+   * {@code false}).
+   *
+   * @return {@code true} if the table should be created when missing
+   */
+  public boolean isCreateTableIfNotExists() {
+    return createTableIfNotExists;
+  }
+
+  /**
+   * Returns the subscription poll interval in milliseconds (default {@value
+   * #DEFAULT_POLL_INTERVAL_MS}).
+   *
+   * @return the poll interval in milliseconds
+   */
+  public long getPollIntervalMs() {
+    return pollIntervalMs;
+  }
+
+  /**
+   * Returns default options.
+   *
+   * @return a {@code PostgresSessionStoreOptions} with all defaults
+   */
+  public static PostgresSessionStoreOptions defaults() {
+    return builder().build();
+  }
+
+  /**
+   * Creates a new builder.
+   *
+   * @return a new builder
+   */
+  public static Builder builder() {
+    return new Builder();
+  }
+
+  /** Builder for {@link PostgresSessionStoreOptions}. */
+  public static final class Builder {
+    private String tableName = DEFAULT_TABLE;
+    private int checkpointInterval = DEFAULT_CHECKPOINT_INTERVAL;
+    private int shardSize = DEFAULT_SHARD_SIZE;
+    private Function snapshotPathPrefix = o -> "global";
+    private boolean createTableIfNotExists = false;
+    private long pollIntervalMs = DEFAULT_POLL_INTERVAL_MS;
+
+    private Builder() {}
+
+    /**
+     * Sets the PostgreSQL table name.
+     *
+     * @param tableName the table name
+     * @return this builder
+     */
+    public Builder tableName(String tableName) {
+      this.tableName = tableName;
+      return this;
+    }
+
+    /**
+     * Sets the number of turns between full checkpoints.
+     *
+     * @param checkpointInterval the checkpoint interval (must be {@code >= 1})
+     * @return this builder
+     */
+    public Builder checkpointInterval(int checkpointInterval) {
+      this.checkpointInterval = checkpointInterval;
+      return this;
+    }
+
+    /**
+     * Sets the shard size in bytes for checkpoint state.
+     *
+     * @param shardSize the shard size in bytes (must be {@code >= 1})
+     * @return this builder
+     */
+    public Builder shardSize(int shardSize) {
+      this.shardSize = shardSize;
+      return this;
+    }
+
+    /**
+     * Sets the function that derives the per-tenant prefix.
+     *
+     * @param snapshotPathPrefix the prefix function
+     * @return this builder
+     */
+    public Builder snapshotPathPrefix(Function snapshotPathPrefix) {
+      this.snapshotPathPrefix = snapshotPathPrefix;
+      return this;
+    }
+
+    /**
+     * Sets whether to create the table on first use if it does not exist.
+     *
+     * @param createTableIfNotExists whether to create the table when missing
+     * @return this builder
+     */
+    public Builder createTableIfNotExists(boolean createTableIfNotExists) {
+      this.createTableIfNotExists = createTableIfNotExists;
+      return this;
+    }
+
+    /**
+     * Sets the subscription poll interval in milliseconds.
+     *
+     * @param pollIntervalMs the poll interval (must be {@code >= 1})
+     * @return this builder
+     */
+    public Builder pollIntervalMs(long pollIntervalMs) {
+      this.pollIntervalMs = pollIntervalMs;
+      return this;
+    }
+
+    /**
+     * Builds a new {@code PostgresSessionStoreOptions}.
+     *
+     * @return a new options instance
+     */
+    public PostgresSessionStoreOptions build() {
+      if (tableName == null || tableName.isBlank()) {
+        throw new IllegalArgumentException("tableName must be non-empty");
+      }
+      if (checkpointInterval < 1) {
+        throw new IllegalArgumentException("checkpointInterval must be >= 1");
+      }
+      if (shardSize < 1) {
+        throw new IllegalArgumentException("shardSize must be >= 1");
+      }
+      if (snapshotPathPrefix == null) {
+        throw new IllegalArgumentException("snapshotPathPrefix must be non-null");
+      }
+      if (pollIntervalMs < 1) {
+        throw new IllegalArgumentException("pollIntervalMs must be >= 1");
+      }
+      return new PostgresSessionStoreOptions(this);
+    }
+  }
+}
diff --git a/plugins/postgresql/src/main/java/com/google/genkit/plugins/postgresql/session/package-info.java b/plugins/postgresql/src/main/java/com/google/genkit/plugins/postgresql/session/package-info.java
new file mode 100644
index 000000000..833aed206
--- /dev/null
+++ b/plugins/postgresql/src/main/java/com/google/genkit/plugins/postgresql/session/package-info.java
@@ -0,0 +1,27 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+/**
+ * PostgreSQL-backed agent session persistence.
+ *
+ * 
{@link com.google.genkit.plugins.postgresql.session.PostgresSessionStore} implements the
+ * Genkit {@code SessionStore} contract using the sharded checkpoint + RFC-6902 diff + pointer
+ * layout shared with the Firestore, DynamoDB, and Cosmos DB backends. Construct it directly from a
+ * {@code javax.sql.DataSource} and pass it to an agent via {@code AgentConfig.store(...)}.
+ */
+package com.google.genkit.plugins.postgresql.session;
diff --git a/plugins/postgresql/src/test/java/com/google/genkit/plugins/postgresql/session/PostgresSessionStoreOptionsTest.java b/plugins/postgresql/src/test/java/com/google/genkit/plugins/postgresql/session/PostgresSessionStoreOptionsTest.java
new file mode 100644
index 000000000..e365cac5b
--- /dev/null
+++ b/plugins/postgresql/src/test/java/com/google/genkit/plugins/postgresql/session/PostgresSessionStoreOptionsTest.java
@@ -0,0 +1,76 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package com.google.genkit.plugins.postgresql.session;
+
+import static org.junit.jupiter.api.Assertions.assertEquals;
+import static org.junit.jupiter.api.Assertions.assertFalse;
+import static org.junit.jupiter.api.Assertions.assertThrows;
+
+import com.google.genkit.ai.agent.SessionStoreOptions;
+import org.junit.jupiter.api.Test;
+
+/** Unit tests for {@link PostgresSessionStoreOptions}. */
+class PostgresSessionStoreOptionsTest {
+
+  @Test
+  void defaultsAreSane() {
+    PostgresSessionStoreOptions o = PostgresSessionStoreOptions.defaults();
+    assertEquals("genkit_sessions", o.getTableName());
+    assertEquals(25, o.getCheckpointInterval());
+    assertEquals(1024 * 1024, o.getShardSize());
+    assertEquals("global", o.getSnapshotPathPrefix().apply(SessionStoreOptions.empty()));
+    assertFalse(o.isCreateTableIfNotExists());
+    assertEquals(2000L, o.getPollIntervalMs());
+  }
+
+  @Test
+  void customBuilder() {
+    PostgresSessionStoreOptions o =
+        PostgresSessionStoreOptions.builder()
+            .tableName("my_sessions")
+            .checkpointInterval(10)
+            .shardSize(4096)
+            .snapshotPathPrefix(so -> "tenant-1")
+            .createTableIfNotExists(true)
+            .pollIntervalMs(500)
+            .build();
+    assertEquals("my_sessions", o.getTableName());
+    assertEquals(10, o.getCheckpointInterval());
+    assertEquals(4096, o.getShardSize());
+    assertEquals("tenant-1", o.getSnapshotPathPrefix().apply(SessionStoreOptions.empty()));
+    assertEquals(true, o.isCreateTableIfNotExists());
+    assertEquals(500L, o.getPollIntervalMs());
+  }
+
+  @Test
+  void builderValidates() {
+    assertThrows(
+        IllegalArgumentException.class,
+        () -> PostgresSessionStoreOptions.builder().tableName("").build());
+    assertThrows(
+        IllegalArgumentException.class,
+        () -> PostgresSessionStoreOptions.builder().checkpointInterval(0).build());
+    assertThrows(
+        IllegalArgumentException.class,
+        () -> PostgresSessionStoreOptions.builder().shardSize(0).build());
+    assertThrows(
+        IllegalArgumentException.class,
+        () -> PostgresSessionStoreOptions.builder().pollIntervalMs(0).build());
+  }
+}
diff --git a/plugins/postgresql/src/test/java/com/google/genkit/plugins/postgresql/session/PostgresSessionStoreTest.java b/plugins/postgresql/src/test/java/com/google/genkit/plugins/postgresql/session/PostgresSessionStoreTest.java
new file mode 100644
index 000000000..53176240b
--- /dev/null
+++ b/plugins/postgresql/src/test/java/com/google/genkit/plugins/postgresql/session/PostgresSessionStoreTest.java
@@ -0,0 +1,212 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package com.google.genkit.plugins.postgresql.session;
+
+import static org.junit.jupiter.api.Assertions.assertEquals;
+import static org.junit.jupiter.api.Assertions.assertNotNull;
+import static org.junit.jupiter.api.Assertions.assertNull;
+import static org.junit.jupiter.api.Assertions.assertThrows;
+import static org.junit.jupiter.api.Assumptions.assumeTrue;
+
+import com.google.genkit.ai.Message;
+import com.google.genkit.ai.agent.GetSnapshotOptions;
+import com.google.genkit.ai.agent.SessionSnapshot;
+import com.google.genkit.ai.agent.SessionState;
+import com.google.genkit.ai.agent.SessionStoreOptions;
+import com.google.genkit.ai.agent.SnapshotStatus;
+import com.google.genkit.core.GenkitException;
+import java.sql.Connection;
+import java.sql.Statement;
+import java.util.HashMap;
+import java.util.List;
+import java.util.Map;
+import java.util.UUID;
+import org.junit.jupiter.api.AfterEach;
+import org.junit.jupiter.api.BeforeEach;
+import org.junit.jupiter.api.Test;
+import org.postgresql.ds.PGSimpleDataSource;
+
+/**
+ * Tests for {@link PostgresSessionStore}.
+ *
+ * 
Integration tests are gated on the {@code POSTGRES_URL} environment variable (e.g. a local
+ * Postgres started with Docker). {@code POSTGRES_USER} and {@code POSTGRES_PASSWORD} default to
+ * {@code postgres}. When {@code POSTGRES_URL} is unset the tests are skipped via {@link
+ * org.junit.jupiter.api.Assumptions}.
+ */
+class PostgresSessionStoreTest {
+
+  private static final String URL = System.getenv("POSTGRES_URL");
+  private static final String USER = envOrDefault("POSTGRES_USER", "postgres");
+  private static final String PASSWORD = envOrDefault("POSTGRES_PASSWORD", "postgres");
+
+  private PGSimpleDataSource dataSource;
+  private String table;
+  private PostgresSessionStore> store;
+
+  private static String envOrDefault(String name, String fallback) {
+    String v = System.getenv(name);
+    return (v != null && !v.isBlank()) ? v : fallback;
+  }
+
+  private static boolean configured() {
+    return URL != null && !URL.isEmpty();
+  }
+
+  @BeforeEach
+  void setUp() {
+    if (!configured()) {
+      return; // integration tests skip via assumeTrue
+    }
+    dataSource = new PGSimpleDataSource();
+    dataSource.setUrl(URL);
+    dataSource.setUser(USER);
+    dataSource.setPassword(PASSWORD);
+    table = "genkit_sessions_test_" + UUID.randomUUID().toString().replace("-", "");
+    store =
+        new PostgresSessionStore<>(
+            dataSource,
+            PostgresSessionStoreOptions.builder()
+                .tableName(table)
+                .checkpointInterval(3)
+                .createTableIfNotExists(true)
+                .build());
+  }
+
+  @AfterEach
+  void tearDown() throws Exception {
+    if (dataSource != null && table != null) {
+      try (Connection conn = dataSource.getConnection();
+          Statement stmt = conn.createStatement()) {
+        stmt.execute("DROP TABLE IF EXISTS \"" + table + "\"");
+      }
+    }
+  }
+
+  private static SessionSnapshot> snapshotWithState(
+      String sessionId, String parentId, Map custom) {
+    SessionState> state =
+        SessionState.>builder()
+            .sessionId(sessionId)
+            .messages(List.of(Message.user("hello")))
+            .custom(custom)
+            .build();
+    return SessionSnapshot.>builder()
+        .sessionId(sessionId)
+        .parentId(parentId)
+        .status(SnapshotStatus.COMPLETED)
+        .state(state)
+        .build();
+  }
+
+  @Test
+  void saveThenGetBySnapshotIdRoundTrips() {
+    assumeTrue(configured());
+    String sessionId = "s-" + UUID.randomUUID();
+    Map custom = new HashMap<>();
+    custom.put("count", 1);
+
+    String id =
+        store.saveSnapshot(
+            null, e -> snapshotWithState(sessionId, null, custom), SessionStoreOptions.empty());
+    assertNotNull(id);
+
+    SessionSnapshot> got =
+        store.getSnapshot(GetSnapshotOptions.builder().snapshotId(id).build());
+    assertNotNull(got);
+    assertEquals(sessionId, got.getSessionId());
+    assertEquals(SnapshotStatus.COMPLETED, got.getStatus());
+    assertEquals(1, got.getState().getMessages().size());
+    assertEquals(1, ((Number) got.getState().getCustom().get("count")).intValue());
+  }
+
+  @Test
+  void getBySessionIdReturnsLeaf() {
+    assumeTrue(configured());
+    String sessionId = "s-" + UUID.randomUUID();
+
+    Map c1 = new HashMap<>();
+    c1.put("count", 1);
+    String id1 =
+        store.saveSnapshot(
+            null, e -> snapshotWithState(sessionId, null, c1), SessionStoreOptions.empty());
+    Map c2 = new HashMap<>();
+    c2.put("count", 2);
+    String id2 =
+        store.saveSnapshot(
+            null, e -> snapshotWithState(sessionId, id1, c2), SessionStoreOptions.empty());
+
+    SessionSnapshot> latest =
+        store.getSnapshot(GetSnapshotOptions.builder().sessionId(sessionId).build());
+    assertNotNull(latest);
+    assertEquals(id2, latest.getSnapshotId());
+    assertEquals(2, ((Number) latest.getState().getCustom().get("count")).intValue());
+  }
+
+  @Test
+  void diffThenCheckpointReconstructs() {
+    assumeTrue(configured());
+    // checkpointInterval is 3; save 5 turns across checkpoint boundaries and confirm the leaf
+    // reconstructs correctly (checkpoint shards + segment-path diffs).
+    String sessionId = "s-" + UUID.randomUUID();
+    String parent = null;
+    String lastId = null;
+    for (int i = 1; i <= 5; i++) {
+      Map c = new HashMap<>();
+      c.put("count", i);
+      final String p = parent;
+      lastId =
+          store.saveSnapshot(
+              null, e -> snapshotWithState(sessionId, p, c), SessionStoreOptions.empty());
+      parent = lastId;
+    }
+    SessionSnapshot> got =
+        store.getSnapshot(GetSnapshotOptions.builder().snapshotId(lastId).build());
+    assertNotNull(got);
+    assertEquals(5, ((Number) got.getState().getCustom().get("count")).intValue());
+  }
+
+  @Test
+  void rejectsEmptySessionId() {
+    assumeTrue(configured());
+    GenkitException ex =
+        assertThrows(
+            GenkitException.class,
+            () ->
+                store.saveSnapshot(
+                    null,
+                    e -> snapshotWithState("", null, new HashMap<>()),
+                    SessionStoreOptions.empty()));
+    assertEquals("INVALID_ARGUMENT", ex.getErrorCode());
+  }
+
+  @Test
+  void mutatorNullIsNoOp() {
+    assumeTrue(configured());
+    assertNull(store.saveSnapshot(null, e -> null, SessionStoreOptions.empty()));
+  }
+
+  @Test
+  void getUnknownSessionReturnsNull() {
+    assumeTrue(configured());
+    assertNull(
+        store.getSnapshot(
+            GetSnapshotOptions.builder().sessionId("s-" + UUID.randomUUID()).build()));
+  }
+}
diff --git a/plugins/qdrant/pom.xml b/plugins/qdrant/pom.xml
new file mode 100644
index 000000000..c7fd86ca4
--- /dev/null
+++ b/plugins/qdrant/pom.xml
@@ -0,0 +1,78 @@
+
+
+
+    4.0.0
+
+    
+        com.google.genkit
+        genkit-parent
+        1.0.0-SNAPSHOT
+        ../../pom.xml
+    
+
+    genkit-plugin-qdrant
+    jar
+    Genkit Qdrant Plugin
+    Qdrant vector database integration for Genkit - indexer and retriever for RAG workflows
+
+    
+        false
+    
+
+    
+        
+        
+            com.google.genkit
+            genkit-core
+            ${project.version}
+        
+        
+            com.google.genkit
+            genkit-ai
+            ${project.version}
+        
+
+        
+        
+            com.fasterxml.jackson.core
+            jackson-databind
+        
+
+        
+        
+            org.slf4j
+            slf4j-api
+        
+
+        
+        
+            org.junit.jupiter
+            junit-jupiter
+            test
+        
+        
+            org.mockito
+            mockito-core
+            test
+        
+    
+
diff --git a/plugins/qdrant/src/main/java/com/google/genkit/plugins/qdrant/QdrantCollectionConfig.java b/plugins/qdrant/src/main/java/com/google/genkit/plugins/qdrant/QdrantCollectionConfig.java
new file mode 100644
index 000000000..108e2decc
--- /dev/null
+++ b/plugins/qdrant/src/main/java/com/google/genkit/plugins/qdrant/QdrantCollectionConfig.java
@@ -0,0 +1,258 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package com.google.genkit.plugins.qdrant;
+
+import java.util.HashMap;
+import java.util.Map;
+
+/**
+ * Configuration for a single Qdrant collection managed by {@link QdrantPlugin}.
+ *
+ * 
Each config registers a retriever and indexer named {@code qdrant/}.
+ */
+public final class QdrantCollectionConfig {
+
+  /** Distance function used by the Qdrant collection. */
+  public enum Distance {
+    COSINE("Cosine"),
+    EUCLIDEAN("Euclid"),
+    DOT_PRODUCT("Dot");
+
+    private final String value;
+
+    Distance(String value) {
+      this.value = value;
+    }
+
+    /**
+     * Returns the Qdrant distance name.
+     *
+     * @return the distance name
+     */
+    public String getValue() {
+      return value;
+    }
+  }
+
+  private final String collectionName;
+  private final String embedderName;
+  private final int dimension;
+  private final Distance distance;
+  private final String textPayloadKey;
+  private final boolean createCollectionIfNotExists;
+  private final Map additionalMetadata;
+
+  private QdrantCollectionConfig(Builder builder) {
+    this.collectionName = builder.collectionName;
+    this.embedderName = builder.embedderName;
+    this.dimension = builder.dimension;
+    this.distance = builder.distance;
+    this.textPayloadKey = builder.textPayloadKey;
+    this.createCollectionIfNotExists = builder.createCollectionIfNotExists;
+    this.additionalMetadata = new HashMap<>(builder.additionalMetadata);
+  }
+
+  /**
+   * Returns the Qdrant collection name.
+   *
+   * @return the collection name
+   */
+  public String getCollectionName() {
+    return collectionName;
+  }
+
+  /**
+   * Returns the name of the embedder used to vectorize documents and queries.
+   *
+   * @return the embedder name
+   */
+  public String getEmbedderName() {
+    return embedderName;
+  }
+
+  /**
+   * Returns the embedding dimension (default {@code 768}).
+   *
+   * @return the embedding dimension
+   */
+  public int getDimension() {
+    return dimension;
+  }
+
+  /**
+   * Returns the distance function (default {@link Distance#COSINE}).
+   *
+   * @return the distance function
+   */
+  public Distance getDistance() {
+    return distance;
+  }
+
+  /**
+   * Returns the payload key that stores the document text (default {@code text}).
+   *
+   * @return the text payload key
+   */
+  public String getTextPayloadKey() {
+    return textPayloadKey;
+  }
+
+  /**
+   * Returns whether to create the collection on first use if it does not exist (default {@code
+   * true}).
+   *
+   * @return {@code true} if the collection should be created when missing
+   */
+  public boolean isCreateCollectionIfNotExists() {
+    return createCollectionIfNotExists;
+  }
+
+  /**
+   * Returns additional metadata merged into every indexed document's payload.
+   *
+   * @return the additional metadata
+   */
+  public Map getAdditionalMetadata() {
+    return additionalMetadata;
+  }
+
+  /**
+   * Creates a new builder.
+   *
+   * @return a new builder
+   */
+  public static Builder builder() {
+    return new Builder();
+  }
+
+  /** Builder for {@link QdrantCollectionConfig}. */
+  public static final class Builder {
+    private String collectionName;
+    private String embedderName;
+    private int dimension = 768;
+    private Distance distance = Distance.COSINE;
+    private String textPayloadKey = "text";
+    private boolean createCollectionIfNotExists = true;
+    private final Map additionalMetadata = new HashMap<>();
+
+    private Builder() {}
+
+    /**
+     * Sets the collection name.
+     *
+     * @param collectionName the collection name
+     * @return this builder
+     */
+    public Builder collectionName(String collectionName) {
+      this.collectionName = collectionName;
+      return this;
+    }
+
+    /**
+     * Sets the embedder name.
+     *
+     * @param embedderName the embedder name
+     * @return this builder
+     */
+    public Builder embedderName(String embedderName) {
+      this.embedderName = embedderName;
+      return this;
+    }
+
+    /**
+     * Sets the embedding dimension.
+     *
+     * @param dimension the embedding dimension (must be {@code >= 1})
+     * @return this builder
+     */
+    public Builder dimension(int dimension) {
+      this.dimension = dimension;
+      return this;
+    }
+
+    /**
+     * Sets the distance function.
+     *
+     * @param distance the distance function
+     * @return this builder
+     */
+    public Builder distance(Distance distance) {
+      this.distance = distance;
+      return this;
+    }
+
+    /**
+     * Sets the payload key that stores the document text.
+     *
+     * @param textPayloadKey the text payload key
+     * @return this builder
+     */
+    public Builder textPayloadKey(String textPayloadKey) {
+      this.textPayloadKey = textPayloadKey;
+      return this;
+    }
+
+    /**
+     * Sets whether to create the collection on first use if it does not exist.
+     *
+     * @param createCollectionIfNotExists whether to create the collection when missing
+     * @return this builder
+     */
+    public Builder createCollectionIfNotExists(boolean createCollectionIfNotExists) {
+      this.createCollectionIfNotExists = createCollectionIfNotExists;
+      return this;
+    }
+
+    /**
+     * Adds a metadata entry merged into every indexed document's payload.
+     *
+     * @param key the metadata key
+     * @param value the metadata value
+     * @return this builder
+     */
+    public Builder addAdditionalMetadata(String key, Object value) {
+      this.additionalMetadata.put(key, value);
+      return this;
+    }
+
+    /**
+     * Builds a new {@code QdrantCollectionConfig}.
+     *
+     * @return a new config instance
+     */
+    public QdrantCollectionConfig build() {
+      if (collectionName == null || collectionName.isBlank()) {
+        throw new IllegalArgumentException("collectionName must be non-empty");
+      }
+      if (embedderName == null || embedderName.isBlank()) {
+        throw new IllegalArgumentException("embedderName must be non-empty");
+      }
+      if (dimension < 1) {
+        throw new IllegalArgumentException("dimension must be >= 1");
+      }
+      if (distance == null) {
+        throw new IllegalArgumentException("distance must be non-null");
+      }
+      if (textPayloadKey == null || textPayloadKey.isBlank()) {
+        throw new IllegalArgumentException("textPayloadKey must be non-empty");
+      }
+      return new QdrantCollectionConfig(this);
+    }
+  }
+}
diff --git a/plugins/qdrant/src/main/java/com/google/genkit/plugins/qdrant/QdrantPlugin.java b/plugins/qdrant/src/main/java/com/google/genkit/plugins/qdrant/QdrantPlugin.java
new file mode 100644
index 000000000..33268895d
--- /dev/null
+++ b/plugins/qdrant/src/main/java/com/google/genkit/plugins/qdrant/QdrantPlugin.java
@@ -0,0 +1,176 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package com.google.genkit.plugins.qdrant;
+
+import com.google.genkit.ai.Embedder;
+import com.google.genkit.core.Action;
+import com.google.genkit.core.ActionType;
+import com.google.genkit.core.Plugin;
+import com.google.genkit.core.Registry;
+import java.util.ArrayList;
+import java.util.List;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+/**
+ * Qdrant vector database plugin for Genkit.
+ *
+ * 
Registers a retriever and indexer named {@code qdrant/} for each configured
+ * collection, talking to a Qdrant server over its REST API.
+ *
+ * 
Example usage:
+ *
+ * 
{@code
+ * Genkit genkit = Genkit.builder()
+ *     .plugin(GoogleGenAIPlugin.create(apiKey))
+ *     .plugin(
+ *         QdrantPlugin.builder()
+ *             .url("http://localhost:6333")
+ *             .addCollection(
+ *                 QdrantCollectionConfig.builder()
+ *                     .collectionName("films")
+ *                     .embedderName("googleai/gemini-embedding-001")
+ *                     .dimension(768)
+ *                     .build())
+ *             .build())
+ *     .build();
+ * }

+ */
+public final class QdrantPlugin implements Plugin {
+
+  /** The plugin name; used as the {@code qdrant/...} action prefix. */
+  public static final String PLUGIN_NAME = "qdrant";
+
+  /** Default Qdrant server URL. */
+  public static final String DEFAULT_URL = "http://localhost:6333";
+
+  private static final Logger logger = LoggerFactory.getLogger(QdrantPlugin.class);
+
+  private final String url;
+  private final String apiKey;
+  private final List collectionConfigs;
+
+  private QdrantPlugin(Builder builder) {
+    this.url = builder.url;
+    this.apiKey = builder.apiKey;
+    this.collectionConfigs = new ArrayList<>(builder.collectionConfigs);
+  }
+
+  /**
+   * Creates a new builder.
+   *
+   * @return a new builder
+   */
+  public static Builder builder() {
+    return new Builder();
+  }
+
+  @Override
+  public String getName() {
+    return PLUGIN_NAME;
+  }
+
+  @Override
+  public List> init() {
+    throw new IllegalStateException(
+        "QdrantPlugin requires a Registry to resolve embedders. Use init(registry) instead.");
+  }
+
+  @Override
+  public List> init(Registry registry) {
+    List> actions = new ArrayList<>();
+    for (QdrantCollectionConfig config : collectionConfigs) {
+      String embedderKey = ActionType.EMBEDDER.keyFromName(config.getEmbedderName());
+      Action embedderAction = registry.lookupAction(embedderKey);
+      if (embedderAction == null) {
+        throw new IllegalStateException(
+            "Embedder not found: "
+                + config.getEmbedderName()
+                + ". Make sure the embedder plugin is registered before QdrantPlugin.");
+      }
+      if (!(embedderAction instanceof Embedder embedder)) {
+        throw new IllegalStateException(
+            "Action " + config.getEmbedderName() + " is not an Embedder");
+      }
+
+      QdrantVectorStore store = new QdrantVectorStore(url, apiKey, config, embedder);
+      actions.add(store.createRetriever());
+      actions.add(store.createIndexer());
+      logger.info("Registered Qdrant vector store: {}/{}", PLUGIN_NAME, config.getCollectionName());
+    }
+    return actions;
+  }
+
+  /** Builder for {@link QdrantPlugin}. */
+  public static final class Builder {
+    private String url = DEFAULT_URL;
+    private String apiKey;
+    private final List collectionConfigs = new ArrayList<>();
+
+    private Builder() {}
+
+    /**
+     * Sets the Qdrant server URL (default {@value #DEFAULT_URL}).
+     *
+     * @param url the server URL
+     * @return this builder
+     */
+    public Builder url(String url) {
+      this.url = url;
+      return this;
+    }
+
+    /**
+     * Sets the Qdrant API key (optional; required for Qdrant Cloud).
+     *
+     * @param apiKey the API key
+     * @return this builder
+     */
+    public Builder apiKey(String apiKey) {
+      this.apiKey = apiKey;
+      return this;
+    }
+
+    /**
+     * Adds a collection configuration.
+     *
+     * @param config the collection configuration
+     * @return this builder
+     */
+    public Builder addCollection(QdrantCollectionConfig config) {
+      this.collectionConfigs.add(config);
+      return this;
+    }
+
+    /**
+     * Builds the plugin.
+     *
+     * @return a new {@code QdrantPlugin}
+     */
+    public QdrantPlugin build() {
+      if (url == null || url.isBlank()) {
+        throw new IllegalStateException("url is required");
+      }
+      if (collectionConfigs.isEmpty()) {
+        throw new IllegalStateException("At least one collection configuration is required");
+      }
+      return new QdrantPlugin(this);
+    }
+  }
+}
diff --git a/plugins/qdrant/src/main/java/com/google/genkit/plugins/qdrant/QdrantVectorStore.java b/plugins/qdrant/src/main/java/com/google/genkit/plugins/qdrant/QdrantVectorStore.java
new file mode 100644
index 000000000..6e7dfc73a
--- /dev/null
+++ b/plugins/qdrant/src/main/java/com/google/genkit/plugins/qdrant/QdrantVectorStore.java
@@ -0,0 +1,333 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package com.google.genkit.plugins.qdrant;
+
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.fasterxml.jackson.databind.node.ArrayNode;
+import com.fasterxml.jackson.databind.node.ObjectNode;
+import com.google.genkit.ai.Document;
+import com.google.genkit.ai.EmbedRequest;
+import com.google.genkit.ai.EmbedResponse;
+import com.google.genkit.ai.Embedder;
+import com.google.genkit.ai.Indexer;
+import com.google.genkit.ai.IndexerRequest;
+import com.google.genkit.ai.IndexerResponse;
+import com.google.genkit.ai.Retriever;
+import com.google.genkit.ai.RetrieverRequest;
+import com.google.genkit.ai.RetrieverResponse;
+import com.google.genkit.core.ActionContext;
+import java.net.URI;
+import java.net.http.HttpClient;
+import java.net.http.HttpRequest;
+import java.net.http.HttpResponse;
+import java.nio.charset.StandardCharsets;
+import java.util.ArrayList;
+import java.util.HashMap;
+import java.util.List;
+import java.util.Map;
+import java.util.UUID;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+/**
+ * Qdrant vector store backed by the Qdrant REST API.
+ *
+ * 
Indexes documents (embedding + payload holding the text and metadata) into a Qdrant collection
+ * and retrieves the nearest neighbors of a query embedding.
+ */
+public final class QdrantVectorStore {
+
+  private static final Logger logger = LoggerFactory.getLogger(QdrantVectorStore.class);
+  private static final ObjectMapper MAPPER = new ObjectMapper();
+
+  private final HttpClient http = HttpClient.newHttpClient();
+  private final String baseUrl;
+  private final String apiKey;
+  private final QdrantCollectionConfig config;
+  private final Embedder embedder;
+
+  private volatile boolean initialized = false;
+
+  /**
+   * Creates a new store.
+   *
+   * @param baseUrl the Qdrant server base URL (e.g. {@code http://localhost:6333})
+   * @param apiKey the Qdrant API key, or {@code null} when the server requires none
+   * @param config the collection configuration
+   * @param embedder the embedder used to vectorize documents and queries
+   */
+  public QdrantVectorStore(
+      String baseUrl, String apiKey, QdrantCollectionConfig config, Embedder embedder) {
+    this.baseUrl = baseUrl.endsWith("/") ? baseUrl.substring(0, baseUrl.length() - 1) : baseUrl;
+    this.apiKey = apiKey;
+    this.config = config;
+    this.embedder = embedder;
+  }
+
+  /** Creates the retriever action registered by the plugin. */
+  Retriever createRetriever() {
+    String name = QdrantPlugin.PLUGIN_NAME + "/" + config.getCollectionName();
+    return Retriever.builder().name(name).handler(this::retrieve).build();
+  }
+
+  /** Creates the indexer action registered by the plugin. */
+  Indexer createIndexer() {
+    String name = QdrantPlugin.PLUGIN_NAME + "/" + config.getCollectionName();
+    return Indexer.builder().name(name).handler(this::index).build();
+  }
+
+  private synchronized void ensureInitialized() {
+    if (initialized) {
+      return;
+    }
+    if (config.isCreateCollectionIfNotExists() && !collectionExists()) {
+      ObjectNode body = MAPPER.createObjectNode();
+      ObjectNode vectors = body.putObject("vectors");
+      vectors.put("size", resolveDimension());
+      vectors.put("distance", config.getDistance().getValue());
+      send("PUT", "/collections/" + config.getCollectionName(), body);
+      logger.info("Created Qdrant collection {}", config.getCollectionName());
+    }
+    initialized = true;
+  }
+
+  /**
+   * Resolves the vector dimension by probing the embedder, falling back to the configured dimension
+   * if the probe fails. This keeps the created collection in sync with whatever embedding model is
+   * wired in.
+   */
+  private int resolveDimension() {
+    try {
+      return generateEmbedding(null, "genkit dimension probe").size();
+    } catch (RuntimeException e) {
+      logger.debug(
+          "Embedding probe failed; using configured dimension {}: {}",
+          config.getDimension(),
+          e.getMessage());
+      return config.getDimension();
+    }
+  }
+
+  private boolean collectionExists() {
+    try {
+      send("GET", "/collections/" + config.getCollectionName(), null);
+      return true;
+    } catch (NotFoundException e) {
+      return false;
+    }
+  }
+
+  /**
+   * Retrieves documents similar to the query.
+   *
+   * @param context the action context
+   * @param request the retriever request
+   * @return the retriever response with matching documents
+   */
+  public RetrieverResponse retrieve(ActionContext context, RetrieverRequest request) {
+    ensureInitialized();
+    Document queryDoc = request.getQuery();
+    if (queryDoc == null || queryDoc.text() == null || queryDoc.text().isBlank()) {
+      throw new RuntimeException("Query document has no text content");
+    }
+    int topK =
+        request.getOptions() != null && request.getOptions().getK() != null
+            ? request.getOptions().getK()
+            : 10;
+    List queryEmbedding = generateEmbedding(context, queryDoc.text());
+
+    ObjectNode body = MAPPER.createObjectNode();
+    body.set("vector", floatsToArray(queryEmbedding));
+    body.put("limit", topK);
+    body.put("with_payload", true);
+
+    JsonNode resp =
+        send("POST", "/collections/" + config.getCollectionName() + "/points/search", body);
+    List documents = new ArrayList<>();
+    JsonNode result = resp.get("result");
+    if (result != null && result.isArray()) {
+      for (JsonNode point : result) {
+        documents.add(toDocument(point));
+      }
+    }
+    logger.debug(
+        "Retrieved {} documents from collection {}", documents.size(), config.getCollectionName());
+    return new RetrieverResponse(documents);
+  }
+
+  /**
+   * Indexes documents into the collection, generating an embedding for each.
+   *
+   * @param context the action context
+   * @param request the indexer request
+   * @return the indexer response
+   */
+  public IndexerResponse index(ActionContext context, IndexerRequest request) {
+    ensureInitialized();
+    List documents = request.getDocuments();
+    if (documents == null || documents.isEmpty()) {
+      logger.warn("No documents to index");
+      return new IndexerResponse();
+    }
+
+    ObjectNode body = MAPPER.createObjectNode();
+    ArrayNode points = body.putArray("points");
+    for (Document doc : documents) {
+      String content = doc.text() != null ? doc.text() : "";
+      List embedding = generateEmbedding(context, content);
+
+      ObjectNode point = points.addObject();
+      point.put("id", getOrGenerateId(doc));
+      point.set("vector", floatsToArray(embedding));
+
+      ObjectNode payload = point.putObject("payload");
+      payload.put(config.getTextPayloadKey(), content);
+      if (doc.getMetadata() != null) {
+        for (Map.Entry entry : doc.getMetadata().entrySet()) {
+          if (!"id".equals(entry.getKey())) {
+            payload.set(entry.getKey(), MAPPER.valueToTree(entry.getValue()));
+          }
+        }
+      }
+      for (Map.Entry entry : config.getAdditionalMetadata().entrySet()) {
+        payload.set(entry.getKey(), MAPPER.valueToTree(entry.getValue()));
+      }
+    }
+
+    send("PUT", "/collections/" + config.getCollectionName() + "/points?wait=true", body);
+    logger.info(
+        "Indexed {} documents into collection {}", documents.size(), config.getCollectionName());
+    return new IndexerResponse();
+  }
+
+  private Document toDocument(JsonNode point) {
+    Map metadata = new HashMap<>();
+    String content = "";
+    JsonNode payload = point.get("payload");
+    if (payload != null && payload.isObject()) {
+      Map payloadMap = MAPPER.convertValue(payload, Map.class);
+      for (Map.Entry entry : payloadMap.entrySet()) {
+        if (entry.getKey().equals(config.getTextPayloadKey())) {
+          content = entry.getValue() != null ? entry.getValue().toString() : "";
+        } else {
+          metadata.put(entry.getKey(), entry.getValue());
+        }
+      }
+    }
+    if (point.get("id") != null) {
+      metadata.put("id", point.get("id").asText());
+    }
+    if (point.get("score") != null && point.get("score").isNumber()) {
+      metadata.put("score", point.get("score").asDouble());
+    }
+    Document doc = new Document(content);
+    doc.setMetadata(metadata);
+    return doc;
+  }
+
+  private ArrayNode floatsToArray(List values) {
+    ArrayNode arr = MAPPER.createArrayNode();
+    for (float v : values) {
+      arr.add(v);
+    }
+    return arr;
+  }
+
+  private List generateEmbedding(ActionContext ctx, String text) {
+    EmbedResponse response = embedder.run(ctx, new EmbedRequest(List.of(new Document(text))));
+    if (response.getEmbeddings() == null || response.getEmbeddings().isEmpty()) {
+      throw new RuntimeException("Failed to generate embedding for text");
+    }
+    float[] values = response.getEmbeddings().get(0).getValues();
+    List out = new ArrayList<>(values.length);
+    for (float v : values) {
+      out.add(v);
+    }
+    return out;
+  }
+
+  private String getOrGenerateId(Document doc) {
+    if (doc.getMetadata() != null && doc.getMetadata().get("id") != null) {
+      String id = doc.getMetadata().get("id").toString();
+      // Qdrant point ids must be an unsigned integer or a UUID; derive a stable UUID otherwise.
+      try {
+        UUID.fromString(id);
+        return id;
+      } catch (IllegalArgumentException e) {
+        return UUID.nameUUIDFromBytes(id.getBytes(StandardCharsets.UTF_8)).toString();
+      }
+    }
+    return UUID.randomUUID().toString();
+  }
+
+  private JsonNode send(String method, String path, JsonNode body) {
+    try {
+      HttpRequest.Builder builder =
+          HttpRequest.newBuilder()
+              .uri(URI.create(baseUrl + path))
+              .header("Content-Type", "application/json")
+              .header("Accept", "application/json");
+      if (apiKey != null && !apiKey.isBlank()) {
+        builder.header("api-key", apiKey);
+      }
+      if (body != null) {
+        builder.method(
+            method,
+            HttpRequest.BodyPublishers.ofString(
+                MAPPER.writeValueAsString(body), StandardCharsets.UTF_8));
+      } else {
+        builder.method(method, HttpRequest.BodyPublishers.noBody());
+      }
+      HttpResponse response =
+          http.send(builder.build(), HttpResponse.BodyHandlers.ofString());
+      if (response.statusCode() == 404) {
+        throw new NotFoundException(path);
+      }
+      if (response.statusCode() / 100 != 2) {
+        throw new RuntimeException(
+            "Qdrant request "
+                + method
+                + " "
+                + path
+                + " failed ("
+                + response.statusCode()
+                + "): "
+                + response.body());
+      }
+      String responseBody = response.body();
+      if (responseBody == null || responseBody.isBlank()) {
+        return MAPPER.createObjectNode();
+      }
+      return MAPPER.readTree(responseBody);
+    } catch (RuntimeException e) {
+      throw e;
+    } catch (Exception e) {
+      throw new RuntimeException(
+          "Qdrant request " + method + " " + path + " failed: " + e.getMessage(), e);
+    }
+  }
+
+  /** Signals a 404 from the Qdrant API (used to detect a missing collection). */
+  private static final class NotFoundException extends RuntimeException {
+    NotFoundException(String path) {
+      super("Not found: " + path);
+    }
+  }
+}
diff --git a/plugins/qdrant/src/main/java/com/google/genkit/plugins/qdrant/package-info.java b/plugins/qdrant/src/main/java/com/google/genkit/plugins/qdrant/package-info.java
new file mode 100644
index 000000000..87061038e
--- /dev/null
+++ b/plugins/qdrant/src/main/java/com/google/genkit/plugins/qdrant/package-info.java
@@ -0,0 +1,25 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+/**
+ * Qdrant vector database integration for Genkit.
+ *
+ * 
{@link com.google.genkit.plugins.qdrant.QdrantPlugin} registers retrievers and indexers backed
+ * by a Qdrant server (REST API) for RAG workflows.
+ */
+package com.google.genkit.plugins.qdrant;
diff --git a/plugins/qdrant/src/test/java/com/google/genkit/plugins/qdrant/QdrantCollectionConfigTest.java b/plugins/qdrant/src/test/java/com/google/genkit/plugins/qdrant/QdrantCollectionConfigTest.java
new file mode 100644
index 000000000..b582c82e6
--- /dev/null
+++ b/plugins/qdrant/src/test/java/com/google/genkit/plugins/qdrant/QdrantCollectionConfigTest.java
@@ -0,0 +1,78 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package com.google.genkit.plugins.qdrant;
+
+import static org.junit.jupiter.api.Assertions.assertEquals;
+import static org.junit.jupiter.api.Assertions.assertThrows;
+import static org.junit.jupiter.api.Assertions.assertTrue;
+
+import org.junit.jupiter.api.Test;
+
+/** Unit tests for {@link QdrantCollectionConfig}. */
+class QdrantCollectionConfigTest {
+
+  @Test
+  void defaultsAreSane() {
+    QdrantCollectionConfig c =
+        QdrantCollectionConfig.builder().collectionName("films").embedderName("e").build();
+    assertEquals("films", c.getCollectionName());
+    assertEquals(768, c.getDimension());
+    assertEquals(QdrantCollectionConfig.Distance.COSINE, c.getDistance());
+    assertEquals("Cosine", c.getDistance().getValue());
+    assertEquals("text", c.getTextPayloadKey());
+    assertTrue(c.isCreateCollectionIfNotExists());
+  }
+
+  @Test
+  void customBuilder() {
+    QdrantCollectionConfig c =
+        QdrantCollectionConfig.builder()
+            .collectionName("docs")
+            .embedderName("e")
+            .dimension(1536)
+            .distance(QdrantCollectionConfig.Distance.DOT_PRODUCT)
+            .textPayloadKey("content")
+            .createCollectionIfNotExists(false)
+            .addAdditionalMetadata("source", "wiki")
+            .build();
+    assertEquals(1536, c.getDimension());
+    assertEquals("Dot", c.getDistance().getValue());
+    assertEquals("content", c.getTextPayloadKey());
+    assertEquals(false, c.isCreateCollectionIfNotExists());
+    assertEquals("wiki", c.getAdditionalMetadata().get("source"));
+  }
+
+  @Test
+  void builderValidates() {
+    assertThrows(
+        IllegalArgumentException.class,
+        () -> QdrantCollectionConfig.builder().embedderName("e").build());
+    assertThrows(
+        IllegalArgumentException.class,
+        () -> QdrantCollectionConfig.builder().collectionName("c").build());
+    assertThrows(
+        IllegalArgumentException.class,
+        () ->
+            QdrantCollectionConfig.builder()
+                .collectionName("c")
+                .embedderName("e")
+                .dimension(0)
+                .build());
+  }
+}
diff --git a/plugins/qdrant/src/test/java/com/google/genkit/plugins/qdrant/QdrantPluginTest.java b/plugins/qdrant/src/test/java/com/google/genkit/plugins/qdrant/QdrantPluginTest.java
new file mode 100644
index 000000000..fd3482910
--- /dev/null
+++ b/plugins/qdrant/src/test/java/com/google/genkit/plugins/qdrant/QdrantPluginTest.java
@@ -0,0 +1,48 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package com.google.genkit.plugins.qdrant;
+
+import static org.junit.jupiter.api.Assertions.assertEquals;
+import static org.junit.jupiter.api.Assertions.assertThrows;
+
+import org.junit.jupiter.api.Test;
+
+/** Unit tests for {@link QdrantPlugin}. */
+class QdrantPluginTest {
+
+  private static QdrantCollectionConfig config() {
+    return QdrantCollectionConfig.builder().collectionName("films").embedderName("e").build();
+  }
+
+  @Test
+  void getName() {
+    assertEquals("qdrant", QdrantPlugin.builder().addCollection(config()).build().getName());
+  }
+
+  @Test
+  void requiresAtLeastOneCollection() {
+    assertThrows(IllegalStateException.class, () -> QdrantPlugin.builder().build());
+  }
+
+  @Test
+  void initWithoutRegistryThrows() {
+    QdrantPlugin plugin = QdrantPlugin.builder().addCollection(config()).build();
+    assertThrows(IllegalStateException.class, plugin::init);
+  }
+}
diff --git a/plugins/qdrant/src/test/java/com/google/genkit/plugins/qdrant/QdrantVectorStoreTest.java b/plugins/qdrant/src/test/java/com/google/genkit/plugins/qdrant/QdrantVectorStoreTest.java
new file mode 100644
index 000000000..123e9a548
--- /dev/null
+++ b/plugins/qdrant/src/test/java/com/google/genkit/plugins/qdrant/QdrantVectorStoreTest.java
@@ -0,0 +1,110 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package com.google.genkit.plugins.qdrant;
+
+import static org.junit.jupiter.api.Assertions.assertEquals;
+import static org.junit.jupiter.api.Assertions.assertFalse;
+import static org.junit.jupiter.api.Assumptions.assumeTrue;
+
+import com.google.genkit.ai.Document;
+import com.google.genkit.ai.EmbedResponse;
+import com.google.genkit.ai.Embedder;
+import com.google.genkit.ai.EmbedderInfo;
+import com.google.genkit.ai.IndexerRequest;
+import com.google.genkit.ai.RetrieverRequest;
+import com.google.genkit.ai.RetrieverResponse;
+import java.util.ArrayList;
+import java.util.List;
+import java.util.Random;
+import java.util.UUID;
+import org.junit.jupiter.api.BeforeEach;
+import org.junit.jupiter.api.Test;
+
+/**
+ * Integration tests for {@link QdrantVectorStore}, gated on the {@code QDRANT_URL} environment
+ * variable (e.g. a Qdrant server started with Docker). Skipped via {@link
+ * org.junit.jupiter.api.Assumptions} when unset. Uses a deterministic stub embedder so a query
+ * equal to an indexed document's text retrieves that document first.
+ */
+class QdrantVectorStoreTest {
+
+  private static final String URL = System.getenv("QDRANT_URL");
+  private static final String API_KEY = System.getenv("QDRANT_API_KEY");
+  private static final int DIM = 16;
+
+  private QdrantVectorStore store;
+
+  private static boolean configured() {
+    return URL != null && !URL.isEmpty();
+  }
+
+  private static Embedder stubEmbedder() {
+    return new Embedder(
+        "test/stub",
+        new EmbedderInfo(),
+        (ctx, req) -> {
+          List out = new ArrayList<>();
+          for (Document doc : req.getDocuments()) {
+            Random random = new Random(doc.text().hashCode());
+            float[] values = new float[DIM];
+            for (int i = 0; i < DIM; i++) {
+              values[i] = random.nextFloat();
+            }
+            out.add(new EmbedResponse.Embedding(values));
+          }
+          return new EmbedResponse(out);
+        });
+  }
+
+  @BeforeEach
+  void setUp() {
+    if (!configured()) {
+      return;
+    }
+    QdrantCollectionConfig config =
+        QdrantCollectionConfig.builder()
+            .collectionName("genkit_test_" + UUID.randomUUID().toString().replace("-", ""))
+            .embedderName("test/stub")
+            .dimension(DIM)
+            .build();
+    store = new QdrantVectorStore(URL, API_KEY, config, stubEmbedder());
+  }
+
+  @Test
+  void indexThenRetrieveReturnsNearestFirst() {
+    assumeTrue(configured());
+    List docs =
+        List.of(
+            Document.fromText("The Matrix is a sci-fi film about simulated reality."),
+            Document.fromText("The Godfather is a crime film about a mafia family."),
+            Document.fromText("Inception is a sci-fi film about dreams."));
+    store.index(null, new IndexerRequest(docs));
+
+    RetrieverRequest request = new RetrieverRequest(Document.fromText(docs.get(1).text()));
+    RetrieverRequest.RetrieverOptions options = new RetrieverRequest.RetrieverOptions();
+    options.setK(3);
+    request.setOptions(options);
+
+    RetrieverResponse response = store.retrieve(null, request);
+    assertFalse(response.getDocuments().isEmpty());
+    assertEquals(
+        "The Godfather is a crime film about a mafia family.",
+        response.getDocuments().get(0).text());
+  }
+}
diff --git a/plugins/weaviate/README.md b/plugins/weaviate/README.md
index c26ad1ea7..876f7e309 100644
--- a/plugins/weaviate/README.md
+++ b/plugins/weaviate/README.md
@@ -58,7 +58,7 @@ Genkit genkit = Genkit.builder()
     .plugin(WeaviatePlugin.local()
         .addCollection(WeaviateCollectionConfig.builder()
             .name("documents")
-            .embedderName("googleai/text-embedding-004")
+            .embedderName("googleai/gemini-embedding-001")
             .build())
         .build())
     .build();
@@ -76,7 +76,7 @@ Genkit genkit = Genkit.builder()
         .apiKey(System.getenv("WEAVIATE_API_KEY"))
         .addCollection(WeaviateCollectionConfig.builder()
             .name("documents")
-            .embedderName("googleai/text-embedding-004")
+            .embedderName("googleai/gemini-embedding-001")
             .distanceMeasure(WeaviateCollectionConfig.DistanceMeasure.COSINE)
             .createCollectionIfMissing(true)
             .vectorDimension(768)
@@ -137,7 +137,7 @@ Flow ragFlow = genkit.defineFlow("ragQuery",
     
     // Generate answer using context
     ModelResponse response = genkit.generate(GenerateOptions.builder()
-        .model("googleai/gemini-2.0-flash")
+        .model("googleai/gemini-2.5-flash")
         .prompt(question)
         .docs(docs)
         .build());
diff --git a/plugins/weaviate/src/main/java/com/google/genkit/plugins/weaviate/WeaviateCollectionConfig.java b/plugins/weaviate/src/main/java/com/google/genkit/plugins/weaviate/WeaviateCollectionConfig.java
index a9b8f3e1c..5e29e756e 100644
--- a/plugins/weaviate/src/main/java/com/google/genkit/plugins/weaviate/WeaviateCollectionConfig.java
+++ b/plugins/weaviate/src/main/java/com/google/genkit/plugins/weaviate/WeaviateCollectionConfig.java
@@ -28,7 +28,7 @@
  * 
{@code
  * WeaviateCollectionConfig config = WeaviateCollectionConfig.builder()
  *     .name("documents")
- *     .embedderName("googleai/text-embedding-004")
+ *     .embedderName("googleai/gemini-embedding-001")
  *     .distanceMeasure(DistanceMeasure.COSINE)
  *     .createCollectionIfMissing(true)
  *     .build();
@@ -201,7 +201,7 @@ public Builder embedder(Embedder embedder) {
     /**
      * Sets the embedder name to resolve from registry.
      *
-     * @param embedderName the embedder name (e.g., "googleai/text-embedding-004")
+     * @param embedderName the embedder name (e.g., "googleai/gemini-embedding-001")
      * @return this builder
      */
     public Builder embedderName(String embedderName) {
diff --git a/plugins/weaviate/src/main/java/com/google/genkit/plugins/weaviate/WeaviatePlugin.java b/plugins/weaviate/src/main/java/com/google/genkit/plugins/weaviate/WeaviatePlugin.java
index 96630677c..83f0d60ba 100644
--- a/plugins/weaviate/src/main/java/com/google/genkit/plugins/weaviate/WeaviatePlugin.java
+++ b/plugins/weaviate/src/main/java/com/google/genkit/plugins/weaviate/WeaviatePlugin.java
@@ -56,7 +56,7 @@
  *             .addCollection(
  *                 WeaviateCollectionConfig.builder()
  *                     .name("documents")
- *                     .embedderName("googleai/text-embedding-004")
+ *                     .embedderName("googleai/gemini-embedding-001")
  *                     .build())
  *             .build())
  *     .build();
diff --git a/plugins/weaviate/src/main/java/com/google/genkit/plugins/weaviate/package-info.java b/plugins/weaviate/src/main/java/com/google/genkit/plugins/weaviate/package-info.java
index a1d2ae391..03228f0d6 100644
--- a/plugins/weaviate/src/main/java/com/google/genkit/plugins/weaviate/package-info.java
+++ b/plugins/weaviate/src/main/java/com/google/genkit/plugins/weaviate/package-info.java
@@ -40,7 +40,7 @@
  *             .addCollection(
  *                 WeaviateCollectionConfig.builder()
  *                     .name("documents")
- *                     .embedderName("googleai/text-embedding-004")
+ *                     .embedderName("googleai/gemini-embedding-001")
  *                     .build())
  *             .build())
  *     .build();
diff --git a/pom.xml b/pom.xml
index d9d125dd9..e9a79e558 100644
--- a/pom.xml
+++ b/pom.xml
@@ -87,6 +87,9 @@
         plugins/middleware
         plugins/weaviate
         plugins/postgresql
+        plugins/mongodb
+        plugins/chroma
+        plugins/qdrant
         plugins/pinecone
         plugins/evaluators
         plugins/aws-bedrock
@@ -116,11 +119,16 @@
         samples/agents-firestore-session
         samples/agents-dynamodb-session
         samples/agents-cosmos-session
+        samples/agents-postgres-session
+        samples/agents-mongo-session
         samples/spring
         samples/firebase
         samples/weaviate
         samples/postgresql
         samples/pinecone
+        samples/chroma
+        samples/qdrant
+        samples/mongo-vector
         samples/structured-output
         samples/aws-bedrock
         samples/azure-foundry
diff --git a/samples/README.md b/samples/README.md
index f258a0715..0e910b6ef 100644
--- a/samples/README.md
+++ b/samples/README.md
@@ -77,10 +77,15 @@ The Dev UI will be available at `http://localhost:4000` and allows you to:
 | [weaviate](./weaviate) | Weaviate vector database RAG sample | `OPENAI_API_KEY` |
 | [postgresql](./postgresql) | PostgreSQL pgvector RAG sample | `OPENAI_API_KEY` |
 | [pinecone](./pinecone) | Pinecone vector database RAG sample | `OPENAI_API_KEY` + `PINECONE_API_KEY` |
+| [chroma](./chroma) | Chroma vector database RAG sample | `GEMINI_API_KEY` + Chroma |
+| [qdrant](./qdrant) | Qdrant vector database RAG sample | `GEMINI_API_KEY` + Qdrant |
+| [mongo-vector](./mongo-vector) | MongoDB Atlas Vector Search RAG sample | `GEMINI_API_KEY` + MongoDB Atlas |
 | [agents-human-in-the-loop](./agents-human-in-the-loop) | Agent interrupts + human-in-the-loop resume | `GEMINI_API_KEY` |
 | [agents-firestore-session](./agents-firestore-session) | Agent session persistence backed by Firestore | `GEMINI_API_KEY` + Firestore |
 | [agents-dynamodb-session](./agents-dynamodb-session) | Agent session persistence backed by DynamoDB | AWS credentials + DynamoDB |
 | [agents-cosmos-session](./agents-cosmos-session) | Agent session persistence backed by Azure Cosmos DB | Azure credentials + Cosmos DB |
+| [agents-postgres-session](./agents-postgres-session) | Agent session persistence backed by PostgreSQL | `GEMINI_API_KEY` + PostgreSQL |
+| [agents-mongo-session](./agents-mongo-session) | Agent session persistence backed by MongoDB | `GEMINI_API_KEY` + MongoDB |
 
 ## Sample Details
 
diff --git a/samples/agents-mongo-session/README.md b/samples/agents-mongo-session/README.md
new file mode 100644
index 000000000..2e31fd3bf
--- /dev/null
+++ b/samples/agents-mongo-session/README.md
@@ -0,0 +1,65 @@
+# Agents: MongoDB Session Sample
+
+A server-managed assistant agent whose conversation state is persisted in **MongoDB** via `MongoSessionStore`, using a **Gemini** model for generation.
+
+## Prerequisites
+
+- Java 21+ and Maven 3.6+
+- A `GEMINI_API_KEY` (for live model calls) — get one from [Google AI Studio](https://aistudio.google.com/apikey)
+- A reachable MongoDB instance (see the Docker command below)
+
+## Run MongoDB locally with Docker
+
+Start a throwaway MongoDB container:
+
+```bash
+docker run --detach \
+  --name genkit-mongo \
+  --publish 27017:27017 \
+  mongo:8
+
+# wait until it's accepting connections
+until docker exec genkit-mongo mongosh --quiet --eval 'db.runCommand({ ping: 1 })' >/dev/null 2>&1; do sleep 1; done
+echo "mongo ready"
+```
+
+The store creates its database (`genkit`) and collection (`genkit_sessions`) automatically on first write — no setup needed. Stop and remove the container later with:
+
+```bash
+docker rm -f genkit-mongo
+```
+
+## Configure
+
+```bash
+export MONGO_URI=mongodb://localhost:27017
+export GEMINI_API_KEY=
+```
+
+### Inspecting the stored data
+
+```bash
+docker exec -it genkit-mongo mongosh genkit \
+  --quiet --eval 'db.genkit_sessions.find({}, { _id: 1 }).toArray()'
+```
+
+Documents are keyed by `_id` = `::` (prefix defaults to `global`): `SNAP_` (snapshot metadata), `SHARD__` (the state JSON), and `PTR_` (the current-leaf pointer). Each document carries a `version` field used for optimistic concurrency.
+
+## Run
+
+**Serve over HTTP (default):**
+
+```bash
+mvn -q exec:java
+# assistant -> POST http://localhost:8084/assistant
+```
+
+**Genkit Dev UI:**
+
+```bash
+genkit start -- mvn -q exec:java
+```
+
+## Configuration
+
+`MongoSessionStore` uses a sharded checkpoint + diff + pointer layout in a single collection. Tune it with `MongoSessionStoreOptions` (database/collection names, checkpoint interval, shard size — default 1 MiB, under the 16 MB document cap — per-tenant prefix).
diff --git a/samples/agents-mongo-session/pom.xml b/samples/agents-mongo-session/pom.xml
new file mode 100644
index 000000000..4e66b505f
--- /dev/null
+++ b/samples/agents-mongo-session/pom.xml
@@ -0,0 +1,87 @@
+
+
+
+    4.0.0
+
+    
+        com.google.genkit
+        genkit-parent
+        1.0.0-SNAPSHOT
+        ../../pom.xml
+    
+
+    com.google.genkit.samples
+    genkit-sample-agents-mongo-session
+    jar
+    Genkit Agents Mongo Session Sample
+    Sample demonstrating agent session persistence backed by MongoDB
+
+    
+        UTF-8
+        21
+        21
+        1.0.0-SNAPSHOT
+        true
+        true
+    
+
+    
+        
+            com.google.genkit
+            genkit
+            ${genkit.version}
+        
+        
+            com.google.genkit
+            genkit-plugin-google-genai
+            ${genkit.version}
+        
+        
+            com.google.genkit
+            genkit-plugin-mongodb
+            ${genkit.version}
+        
+        
+            com.google.genkit
+            genkit-plugin-jetty
+            ${genkit.version}
+        
+        
+            ch.qos.logback
+            logback-classic
+            1.5.37
+        
+    
+
+    
+        
+            
+                org.codehaus.mojo
+                exec-maven-plugin
+                3.6.3
+                
+                    com.google.genkit.samples.MongoSessionAgentApp
+                
+            
+        
+    
+
diff --git a/samples/agents-mongo-session/run.sh b/samples/agents-mongo-session/run.sh
new file mode 100755
index 000000000..d10fb3e2a
--- /dev/null
+++ b/samples/agents-mongo-session/run.sh
@@ -0,0 +1,7 @@
+#!/bin/bash
+
+# Genkit Sample Runner
+# This script runs the sample application
+
+echo "Building and running the sample..."
+mvn clean compile exec:java
diff --git a/samples/agents-mongo-session/src/main/java/com/google/genkit/samples/MongoSessionAgentApp.java b/samples/agents-mongo-session/src/main/java/com/google/genkit/samples/MongoSessionAgentApp.java
new file mode 100644
index 000000000..7a2e2f347
--- /dev/null
+++ b/samples/agents-mongo-session/src/main/java/com/google/genkit/samples/MongoSessionAgentApp.java
@@ -0,0 +1,119 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package com.google.genkit.samples;
+
+import com.google.genkit.Genkit;
+import com.google.genkit.GenkitOptions;
+import com.google.genkit.agent.AgentConfig;
+import com.google.genkit.ai.agent.SessionStore;
+import com.google.genkit.plugins.googlegenai.GoogleGenAIPlugin;
+import com.google.genkit.plugins.jetty.JettyPlugin;
+import com.google.genkit.plugins.mongodb.session.MongoSessionStore;
+import com.google.genkit.plugins.mongodb.session.MongoSessionStoreOptions;
+import com.mongodb.client.MongoClient;
+import com.mongodb.client.MongoClients;
+import java.util.Map;
+
+/**
+ * Agent session persistence backed by MongoDB.
+ *
+ * 
Defines a server-managed assistant whose conversation snapshots are stored in MongoDB via
+ * {@link MongoSessionStore}. Because the state lives in the database (not in the process), the
+ * conversation survives restarts and can be resumed from any instance.
+ *
+ * 
Point it at a local MongoDB (see the README for a one-line Docker command):
+ *
+ * 
+ *   export MONGO_URI=mongodb://localhost:27017
+ *   export GEMINI_API_KEY=<your-key>
+ * 

+ *
+ * 
Starts Jetty to expose the agent over HTTP and keep the process alive; also discoverable in
+ * the Genkit Dev UI. {@code POST http://localhost:8084/assistant}.
+ */
+public class MongoSessionAgentApp {
+
+  public static void main(String[] args) throws Exception {
+    // ── 1. Build Genkit with the beta agents API enabled ────────────────────
+    Genkit genkit =
+        Genkit.builder()
+            .options(GenkitOptions.builder().experimental(true).devMode(true).build())
+            .plugin(GoogleGenAIPlugin.create())
+            .build();
+
+    // ── 2. Build the MongoDB-backed session store (null if not configured) ───
+    SessionStore> store = buildStore();
+
+    // ── 3. Define a server-managed agent using the store ────────────────────
+    AgentConfig.Builder> cfg =
+        AgentConfig.>builder()
+            .name("assistant")
+            .description("A helpful assistant with MongoDB-backed memory")
+            .system(
+                "You are a helpful assistant. Keep answers concise and remember what the user tells you.")
+            .model("googleai/gemini-2.5-flash");
+    if (store != null) {
+      cfg.store(store);
+    }
+    genkit.beta().defineAgent(cfg.build());
+
+    // ── 4. Serve over HTTP + keep the process alive ─────────────────────────
+    serve(genkit);
+  }
+
+  /**
+   * Builds a MongoDB-backed session store, or returns {@code null} (running client-managed) when
+   * MongoDB is not configured / cannot be reached. Reads {@code MONGO_URI} (connection string, e.g.
+   * {@code mongodb://localhost:27017}).
+   */
+  private static SessionStore> buildStore() {
+    String uri = System.getenv("MONGO_URI");
+    if (uri == null || uri.isBlank()) {
+      System.out.println(
+          "MongoDB not configured (set MONGO_URI); running client-managed. See the README for a"
+              + " one-line Docker command.");
+      return null;
+    }
+    try {
+      MongoClient client = MongoClients.create(uri);
+      return new MongoSessionStore<>(client, MongoSessionStoreOptions.defaults());
+    } catch (Exception e) {
+      System.out.println("MongoDB not reachable (" + e.getMessage() + "); running client-managed.");
+      return null;
+    }
+  }
+
+  /** Starts Jetty to serve the agent over HTTP and blocks until the process is stopped. */
+  private static void serve(Genkit genkit) throws Exception {
+    int port = System.getenv("PORT") != null ? Integer.parseInt(System.getenv("PORT")) : 8084;
+    JettyPlugin jetty = JettyPlugin.create(port);
+    jetty.init(genkit.getRegistry());
+    System.out.println("Serving the assistant on http://localhost:" + port);
+    System.out.println("  assistant -> POST http://localhost:" + port + "/assistant");
+    System.out.println("Tip: run under `genkit start -- mvn -q exec:java` to open the Dev UI.");
+    try {
+      jetty.start(); // blocks until the process is stopped (Ctrl-C)
+    } catch (Exception e) {
+      System.err.println(
+          "ERROR: could not start the HTTP server on port " + port + ": " + e.getMessage());
+      System.err.println("Port " + port + " is likely already in use. Free it or set PORT.");
+      throw e;
+    }
+  }
+}
diff --git a/samples/agents-mongo-session/src/main/resources/logback.xml b/samples/agents-mongo-session/src/main/resources/logback.xml
new file mode 100644
index 000000000..42cdb3d0b
--- /dev/null
+++ b/samples/agents-mongo-session/src/main/resources/logback.xml
@@ -0,0 +1,24 @@
+
+
+    
+    
+        
+            %d{HH:mm:ss.SSS} %-5level %logger{24} - %msg%n
+        
+    
+
+    
+    
+               
+    
+    
+    
+     
+
+    
+    
+
+    
+        
+    
+
diff --git a/samples/agents-postgres-session/README.md b/samples/agents-postgres-session/README.md
new file mode 100644
index 000000000..a6b3d5d26
--- /dev/null
+++ b/samples/agents-postgres-session/README.md
@@ -0,0 +1,70 @@
+# Agents: PostgreSQL Session Sample
+
+A server-managed assistant agent whose conversation state is persisted in **PostgreSQL** via `PostgresSessionStore`, using a **Gemini** model for generation.
+
+## Prerequisites
+
+- Java 21+ and Maven 3.6+
+- A `GEMINI_API_KEY` (for live model calls) — get one from [Google AI Studio](https://aistudio.google.com/apikey)
+- A reachable PostgreSQL instance (see the Docker command below)
+
+## Run PostgreSQL locally with Docker
+
+Start a throwaway Postgres container:
+
+```bash
+docker run --detach \
+  --name genkit-postgres \
+  --env POSTGRES_USER=postgres \
+  --env POSTGRES_PASSWORD=postgres \
+  --env POSTGRES_DB=genkit \
+  --publish 5432:5432 \
+  postgres:18
+
+# wait until it's accepting connections
+until docker exec genkit-postgres pg_isready -U postgres >/dev/null 2>&1; do sleep 1; done
+echo "postgres ready"
+```
+
+The store creates its table (`genkit_sessions`) automatically on first use — no schema setup needed. Stop and remove the container later with:
+
+```bash
+docker rm -f genkit-postgres
+```
+
+## Configure
+
+```bash
+export POSTGRES_URL=jdbc:postgresql://localhost:5432/genkit
+export POSTGRES_USER=postgres       # optional (default: postgres)
+export POSTGRES_PASSWORD=postgres   # optional (default: postgres)
+export GEMINI_API_KEY=
+```
+
+### Inspecting the stored data
+
+```bash
+docker exec -it genkit-postgres psql -U postgres -d genkit \
+  -c "SELECT id FROM genkit_sessions;"
+```
+
+Rows are keyed by `id`: `SNAP_` (snapshot metadata), `SHARD__` (the state JSON), and `PTR_` (the current-leaf pointer). Each row carries a JSONB `doc` payload and a `version` counter used for optimistic concurrency.
+
+## Run
+
+**Serve over HTTP (default):**
+
+```bash
+mvn -q exec:java
+# assistant -> POST http://localhost:8083/assistant
+```
+
+**Genkit Dev UI:**
+
+```bash
+genkit start -- mvn -q exec:java
+```
+
+## Configuration
+
+`PostgresSessionStore` uses a sharded checkpoint + diff + pointer layout in a single table. Tune it with `PostgresSessionStoreOptions` (table name, checkpoint interval, shard size — default 1 MiB — per-tenant prefix, `createTableIfNotExists`).
diff --git a/samples/agents-postgres-session/pom.xml b/samples/agents-postgres-session/pom.xml
new file mode 100644
index 000000000..8b385be51
--- /dev/null
+++ b/samples/agents-postgres-session/pom.xml
@@ -0,0 +1,87 @@
+
+
+
+    4.0.0
+
+    
+        com.google.genkit
+        genkit-parent
+        1.0.0-SNAPSHOT
+        ../../pom.xml
+    
+
+    com.google.genkit.samples
+    genkit-sample-agents-postgres-session
+    jar
+    Genkit Agents Postgres Session Sample
+    Sample demonstrating agent session persistence backed by PostgreSQL
+
+    
+        UTF-8
+        21
+        21
+        1.0.0-SNAPSHOT
+        true
+        true
+    
+
+    
+        
+            com.google.genkit
+            genkit
+            ${genkit.version}
+        
+        
+            com.google.genkit
+            genkit-plugin-google-genai
+            ${genkit.version}
+        
+        
+            com.google.genkit
+            genkit-plugin-postgresql
+            ${genkit.version}
+        
+        
+            com.google.genkit
+            genkit-plugin-jetty
+            ${genkit.version}
+        
+        
+            ch.qos.logback
+            logback-classic
+            1.5.37
+        
+    
+
+    
+        
+            
+                org.codehaus.mojo
+                exec-maven-plugin
+                3.6.3
+                
+                    com.google.genkit.samples.PostgresSessionAgentApp
+                
+            
+        
+    
+
diff --git a/samples/agents-postgres-session/run.sh b/samples/agents-postgres-session/run.sh
new file mode 100755
index 000000000..d10fb3e2a
--- /dev/null
+++ b/samples/agents-postgres-session/run.sh
@@ -0,0 +1,7 @@
+#!/bin/bash
+
+# Genkit Sample Runner
+# This script runs the sample application
+
+echo "Building and running the sample..."
+mvn clean compile exec:java
diff --git a/samples/agents-postgres-session/src/main/java/com/google/genkit/samples/PostgresSessionAgentApp.java b/samples/agents-postgres-session/src/main/java/com/google/genkit/samples/PostgresSessionAgentApp.java
new file mode 100644
index 000000000..bd0f98152
--- /dev/null
+++ b/samples/agents-postgres-session/src/main/java/com/google/genkit/samples/PostgresSessionAgentApp.java
@@ -0,0 +1,135 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package com.google.genkit.samples;
+
+import com.google.genkit.Genkit;
+import com.google.genkit.GenkitOptions;
+import com.google.genkit.agent.AgentConfig;
+import com.google.genkit.ai.agent.SessionStore;
+import com.google.genkit.plugins.googlegenai.GoogleGenAIPlugin;
+import com.google.genkit.plugins.jetty.JettyPlugin;
+import com.google.genkit.plugins.postgresql.session.PostgresSessionStore;
+import com.google.genkit.plugins.postgresql.session.PostgresSessionStoreOptions;
+import java.util.Map;
+import javax.sql.DataSource;
+import org.postgresql.ds.PGSimpleDataSource;
+
+/**
+ * Agent session persistence backed by PostgreSQL.
+ *
+ * 
Defines a server-managed assistant whose conversation snapshots are stored in PostgreSQL via
+ * {@link PostgresSessionStore}. Because the state lives in the database (not in the process), the
+ * conversation survives restarts and can be resumed from any instance.
+ *
+ * 
Point it at a local Postgres (see the README for a one-line Docker command):
+ *
+ * 
+ *   export POSTGRES_URL=jdbc:postgresql://localhost:5432/genkit
+ *   export POSTGRES_USER=postgres
+ *   export POSTGRES_PASSWORD=postgres
+ *   export GEMINI_API_KEY=<your-key>
+ * 

+ *
+ * 
Starts Jetty to expose the agent over HTTP and keep the process alive; also discoverable in
+ * the Genkit Dev UI. {@code POST http://localhost:8083/assistant}.
+ */
+public class PostgresSessionAgentApp {
+
+  public static void main(String[] args) throws Exception {
+    // ── 1. Build Genkit with the beta agents API enabled ────────────────────
+    Genkit genkit =
+        Genkit.builder()
+            .options(GenkitOptions.builder().experimental(true).devMode(true).build())
+            .plugin(GoogleGenAIPlugin.create())
+            .build();
+
+    // ── 2. Build the Postgres-backed session store (null if not configured) ──
+    SessionStore> store = buildStore();
+
+    // ── 3. Define a server-managed agent using the store ────────────────────
+    AgentConfig.Builder> cfg =
+        AgentConfig.>builder()
+            .name("assistant")
+            .description("A helpful assistant with PostgreSQL-backed memory")
+            .system(
+                "You are a helpful assistant. Keep answers concise and remember what the user tells you.")
+            .model("googleai/gemini-2.5-flash");
+    if (store != null) {
+      cfg.store(store);
+    }
+    genkit.beta().defineAgent(cfg.build());
+
+    // ── 4. Serve over HTTP + keep the process alive ─────────────────────────
+    serve(genkit);
+  }
+
+  /**
+   * Builds a PostgreSQL-backed session store, or returns {@code null} (running client-managed) when
+   * PostgreSQL is not configured / cannot be reached. Reads {@code POSTGRES_URL} (JDBC URL, default
+   * {@code jdbc:postgresql://localhost:5432/genkit}), {@code POSTGRES_USER} (default {@code
+   * postgres}), and {@code POSTGRES_PASSWORD} (default {@code postgres}).
+   */
+  private static SessionStore> buildStore() {
+    String url = System.getenv("POSTGRES_URL");
+    if (url == null || url.isBlank()) {
+      System.out.println(
+          "PostgreSQL not configured (set POSTGRES_URL); running client-managed. See the README"
+              + " for a one-line Docker command.");
+      return null;
+    }
+    String user = envOrDefault("POSTGRES_USER", "postgres");
+    String password = envOrDefault("POSTGRES_PASSWORD", "postgres");
+    try {
+      PGSimpleDataSource ds = new PGSimpleDataSource();
+      ds.setUrl(url);
+      ds.setUser(user);
+      ds.setPassword(password);
+      DataSource dataSource = ds;
+      return new PostgresSessionStore<>(
+          dataSource, PostgresSessionStoreOptions.builder().createTableIfNotExists(true).build());
+    } catch (Exception e) {
+      System.out.println(
+          "PostgreSQL not reachable (" + e.getMessage() + "); running client-managed.");
+      return null;
+    }
+  }
+
+  private static String envOrDefault(String name, String fallback) {
+    String v = System.getenv(name);
+    return (v != null && !v.isBlank()) ? v : fallback;
+  }
+
+  /** Starts Jetty to serve the agent over HTTP and blocks until the process is stopped. */
+  private static void serve(Genkit genkit) throws Exception {
+    int port = System.getenv("PORT") != null ? Integer.parseInt(System.getenv("PORT")) : 8083;
+    JettyPlugin jetty = JettyPlugin.create(port);
+    jetty.init(genkit.getRegistry());
+    System.out.println("Serving the assistant on http://localhost:" + port);
+    System.out.println("  assistant -> POST http://localhost:" + port + "/assistant");
+    System.out.println("Tip: run under `genkit start -- mvn -q exec:java` to open the Dev UI.");
+    try {
+      jetty.start(); // blocks until the process is stopped (Ctrl-C)
+    } catch (Exception e) {
+      System.err.println(
+          "ERROR: could not start the HTTP server on port " + port + ": " + e.getMessage());
+      System.err.println("Port " + port + " is likely already in use. Free it or set PORT.");
+      throw e;
+    }
+  }
+}
diff --git a/samples/agents-postgres-session/src/main/resources/logback.xml b/samples/agents-postgres-session/src/main/resources/logback.xml
new file mode 100644
index 000000000..ab38a6f29
--- /dev/null
+++ b/samples/agents-postgres-session/src/main/resources/logback.xml
@@ -0,0 +1,24 @@
+
+
+    
+    
+        
+            %d{HH:mm:ss.SSS} %-5level %logger{24} - %msg%n
+        
+    
+
+    
+    
+               
+    
+    
+    
+     
+
+    
+    
+
+    
+        
+    
+
diff --git a/samples/chroma/README.md b/samples/chroma/README.md
new file mode 100644
index 000000000..8f591e007
--- /dev/null
+++ b/samples/chroma/README.md
@@ -0,0 +1,60 @@
+# Chroma RAG Sample
+
+A Retrieval-Augmented Generation sample that indexes film descriptions into **Chroma** via `ChromaPlugin` and answers questions with a **Gemini** model.
+
+## Prerequisites
+
+- Java 21+ and Maven 3.6+
+- A `GEMINI_API_KEY` — get one from [Google AI Studio](https://aistudio.google.com/apikey)
+- A reachable Chroma server (see the Docker command below)
+
+## Run Chroma locally with Docker
+
+```bash
+docker run --detach \
+  --name genkit-chroma \
+  --publish 8000:8000 \
+  chromadb/chroma
+
+# wait until it's accepting connections
+until curl -sf http://localhost:8000/api/v2/heartbeat >/dev/null 2>&1; do sleep 1; done
+echo "chroma ready"
+```
+
+The collection is created automatically on first use. Stop and remove the container later with:
+
+```bash
+docker rm -f genkit-chroma
+```
+
+## Configure
+
+```bash
+export GEMINI_API_KEY=
+export CHROMA_URL=http://localhost:8000        # optional (default)
+export CHROMA_COLLECTION=genkit_films          # optional (default)
+```
+
+## Run
+
+```bash
+mvn -q exec:java
+```
+
+Then open the Genkit Dev UI at http://localhost:4000 (or run under `genkit start -- mvn -q exec:java`) and exercise the flows:
+
+- `indexDocuments` — index the sample film descriptions
+- `retrieveDocuments` — return the films matching a query
+- `ragQuery` — answer a question using retrieved context
+
+Or via curl:
+
+```bash
+curl -X POST http://localhost:4000/api/flows/indexDocuments -H 'Content-Type: application/json' -d '{}'
+curl -X POST http://localhost:4000/api/flows/ragQuery -H 'Content-Type: application/json' \
+  -d '{"data": "What Christopher Nolan films are mentioned?"}'
+```
+
+## Configuration
+
+`ChromaPlugin` talks to the Chroma v2 REST API. Tune per-collection settings with `ChromaCollectionConfig` (collection name, embedder, distance function, `createCollectionIfNotExists`, additional metadata).
diff --git a/samples/chroma/pom.xml b/samples/chroma/pom.xml
new file mode 100644
index 000000000..17f77abc9
--- /dev/null
+++ b/samples/chroma/pom.xml
@@ -0,0 +1,86 @@
+
+
+
+    4.0.0
+
+    
+        com.google.genkit
+        genkit-parent
+        1.0.0-SNAPSHOT
+        ../../pom.xml
+    
+
+    genkit-sample-chroma
+    jar
+    Genkit Chroma RAG Sample
+    Sample application demonstrating Chroma vector search with Genkit
+
+    
+        true
+        com.google.genkit.samples.chroma.ChromaRAGSample
+    
+
+    
+        
+            com.google.genkit
+            genkit
+            ${project.version}
+        
+        
+            com.google.genkit
+            genkit-plugin-chroma
+            ${project.version}
+        
+        
+            com.google.genkit
+            genkit-plugin-google-genai
+            ${project.version}
+        
+        
+            com.google.genkit
+            genkit-plugin-jetty
+            ${project.version}
+        
+        
+            ch.qos.logback
+            logback-classic
+        
+        
+            io.github.cdimascio
+            dotenv-java
+            3.2.0
+        
+    
+
+    
+        
+            
+                org.codehaus.mojo
+                exec-maven-plugin
+                3.6.3
+                
+                    ${exec.mainClass}
+                
+            
+        
+    
+
diff --git a/samples/chroma/run.sh b/samples/chroma/run.sh
new file mode 100755
index 000000000..d10fb3e2a
--- /dev/null
+++ b/samples/chroma/run.sh
@@ -0,0 +1,7 @@
+#!/bin/bash
+
+# Genkit Sample Runner
+# This script runs the sample application
+
+echo "Building and running the sample..."
+mvn clean compile exec:java
diff --git a/samples/chroma/src/main/java/com/google/genkit/samples/chroma/ChromaRAGSample.java b/samples/chroma/src/main/java/com/google/genkit/samples/chroma/ChromaRAGSample.java
new file mode 100644
index 000000000..ff841d2c0
--- /dev/null
+++ b/samples/chroma/src/main/java/com/google/genkit/samples/chroma/ChromaRAGSample.java
@@ -0,0 +1,163 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package com.google.genkit.samples.chroma;
+
+import com.google.genkit.Genkit;
+import com.google.genkit.GenkitOptions;
+import com.google.genkit.ai.*;
+import com.google.genkit.core.Flow;
+import com.google.genkit.plugins.chroma.ChromaCollectionConfig;
+import com.google.genkit.plugins.chroma.ChromaPlugin;
+import com.google.genkit.plugins.googlegenai.GoogleGenAIPlugin;
+import com.google.genkit.plugins.jetty.JettyPlugin;
+import com.google.genkit.plugins.jetty.JettyPluginOptions;
+import io.github.cdimascio.dotenv.Dotenv;
+import java.util.List;
+import java.util.stream.Collectors;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+/**
+ * Sample application demonstrating Chroma vector search with Genkit.
+ *
+ * 
Indexes a handful of film descriptions into a Chroma collection, retrieves the nearest matches
+ * for a query, and answers questions with a RAG flow. Requires a running Chroma server (see the
+ * README for a one-line Docker command) and a {@code GEMINI_API_KEY}.
+ */
+public class ChromaRAGSample {
+
+  private static final Logger logger = LoggerFactory.getLogger(ChromaRAGSample.class);
+
+  private static final List SAMPLE_DOCUMENTS =
+      List.of(
+          "The Godfather is a 1972 crime film directed by Francis Ford Coppola about the Corleone crime family.",
+          "The Dark Knight is a 2008 superhero film directed by Christopher Nolan featuring Batman against the Joker.",
+          "Pulp Fiction is a 1994 crime film directed by Quentin Tarantino known for its nonlinear narrative.",
+          "Inception is a 2010 sci-fi film directed by Christopher Nolan about dream infiltration.",
+          "The Matrix is a 1999 sci-fi film directed by the Wachowskis exploring simulated reality.",
+          "Forrest Gump is a 1994 drama directed by Robert Zemeckis about a man's extraordinary life.",
+          "Star Wars is a 1977 sci-fi film directed by George Lucas set in a galaxy far, far away.",
+          "The Shawshank Redemption is a 1994 drama about hope and friendship in a prison.");
+
+  private static final String RAG_SYSTEM_PROMPT =
+      """
+      You are a helpful assistant that answers questions based on the provided context documents.
+      Answer only from the context. If the context is insufficient, say so.
+      """;
+
+  public static void main(String[] args) {
+    Dotenv dotenv = Dotenv.configure().ignoreIfMissing().systemProperties().load();
+
+    String geminiApiKey = getEnv(dotenv, "GEMINI_API_KEY");
+    if (geminiApiKey == null) {
+      logger.error("Please set GEMINI_API_KEY in .env file or environment variable");
+      System.exit(1);
+    }
+
+    String chromaUrl = getEnvOrDefault(dotenv, "CHROMA_URL", "http://localhost:8000");
+    String collection = getEnvOrDefault(dotenv, "CHROMA_COLLECTION", "genkit_films");
+
+    logger.info("Starting Chroma RAG Sample (url={}, collection={})", chromaUrl, collection);
+
+    ChromaPlugin chromaPlugin =
+        ChromaPlugin.builder()
+            .url(chromaUrl)
+            .addCollection(
+                ChromaCollectionConfig.builder()
+                    .collectionName(collection)
+                    .embedderName("googleai/gemini-embedding-001")
+                    .distance(ChromaCollectionConfig.Distance.COSINE)
+                    .build())
+            .build();
+
+    JettyPlugin jetty = new JettyPlugin(JettyPluginOptions.builder().port(8080).build());
+    Genkit genkit =
+        Genkit.builder()
+            .options(GenkitOptions.builder().devMode(true).reflectionPort(3100).build())
+            .plugin(GoogleGenAIPlugin.create(geminiApiKey))
+            .plugin(chromaPlugin)
+            .plugin(jetty)
+            .build();
+
+    String action = "chroma/" + collection;
+
+    Flow indexDocumentsFlow =
+        genkit.defineFlow(
+            "indexDocuments",
+            Void.class,
+            String.class,
+            (ctx, input) -> {
+              List documents =
+                  SAMPLE_DOCUMENTS.stream().map(Document::fromText).collect(Collectors.toList());
+              genkit.index(action, documents);
+              return "Successfully indexed " + documents.size() + " documents";
+            });
+
+    @SuppressWarnings("unchecked")
+    Flow, Void> retrieveDocumentsFlow =
+        genkit.defineFlow(
+            "retrieveDocuments",
+            String.class,
+            (Class>) (Class) List.class,
+            (ctx, query) -> {
+              List docs = genkit.retrieve(action, query);
+              return docs.stream().map(Document::text).collect(Collectors.toList());
+            });
+
+    Flow ragQueryFlow =
+        genkit.defineFlow(
+            "ragQuery",
+            String.class,
+            String.class,
+            (ctx, question) -> {
+              List docs = genkit.retrieve(action, question);
+              ModelResponse response =
+                  genkit.generate(
+                      GenerateOptions.builder()
+                          .model("googleai/gemini-2.5-flash")
+                          .system(RAG_SYSTEM_PROMPT)
+                          .prompt(question)
+                          .docs(docs)
+                          .config(GenerationConfig.builder().temperature(0.3).build())
+                          .build());
+              return response.getText();
+            });
+
+    logger.info(
+        "Genkit Chroma RAG Sample started. Flows: indexDocuments, retrieveDocuments, ragQuery");
+    logger.info("Dev UI: http://localhost:4000  |  Reflection: http://localhost:3100");
+
+    try {
+      jetty.start();
+    } catch (Exception e) {
+      logger.error("Failed to start Jetty server", e);
+      System.exit(1);
+    }
+  }
+
+  private static String getEnv(Dotenv dotenv, String name) {
+    String value = dotenv.get(name);
+    return (value != null && !value.isBlank()) ? value : System.getenv(name);
+  }
+
+  private static String getEnvOrDefault(Dotenv dotenv, String name, String defaultValue) {
+    String value = getEnv(dotenv, name);
+    return (value != null && !value.isBlank()) ? value : defaultValue;
+  }
+}
diff --git a/samples/chroma/src/main/resources/logback.xml b/samples/chroma/src/main/resources/logback.xml
new file mode 100644
index 000000000..dff2540b6
--- /dev/null
+++ b/samples/chroma/src/main/resources/logback.xml
@@ -0,0 +1,20 @@
+
+
+    
+        
+            %d{HH:mm:ss.SSS} %-5level %logger{24} - %msg%n
+        
+    
+
+    
+    
+    
+    
+    
+
+    
+
+    
+        
+    
+
diff --git a/samples/firebase/README.md b/samples/firebase/README.md
index 22d2fd653..0c182e528 100644
--- a/samples/firebase/README.md
+++ b/samples/firebase/README.md
@@ -144,7 +144,7 @@ Genkit genkit = Genkit.builder()
         .addRetriever(FirestoreRetrieverConfig.builder()
             .name("my-docs")
             .collection("documents")
-            .embedderName("googleai/text-embedding-004")
+            .embedderName("googleai/gemini-embedding-001")
             .vectorField("embedding")
             .contentField("content")
             .distanceMeasure(FirestoreRetrieverConfig.DistanceMeasure.COSINE)
diff --git a/samples/firebase/src/main/java/com/google/genkit/samples/firebase/FirestoreRAGSample.java b/samples/firebase/src/main/java/com/google/genkit/samples/firebase/FirestoreRAGSample.java
index 90995c998..1992fdd6b 100644
--- a/samples/firebase/src/main/java/com/google/genkit/samples/firebase/FirestoreRAGSample.java
+++ b/samples/firebase/src/main/java/com/google/genkit/samples/firebase/FirestoreRAGSample.java
@@ -120,7 +120,7 @@ public static void main(String[] args) {
                         FirestoreRetrieverConfig.builder()
                             .name("films")
                             .collection("films")
-                            .embedderName("googleai/text-embedding-004")
+                            .embedderName("googleai/gemini-embedding-001")
                             .vectorField("embedding")
                             .contentField("content")
                             .distanceMeasure(FirestoreRetrieverConfig.DistanceMeasure.COSINE)
@@ -198,7 +198,7 @@ public static void main(String[] args) {
               ModelResponse response =
                   genkit.generate(
                       GenerateOptions.builder()
-                          .model("googleai/gemini-2.0-flash")
+                          .model("googleai/gemini-2.5-flash")
                           .system(RAG_SYSTEM_PROMPT)
                           .prompt(question)
                           .docs(docs)
diff --git a/samples/firebase/src/main/java/com/google/genkit/samples/firebase/functions/GeneratePoemFunction.java b/samples/firebase/src/main/java/com/google/genkit/samples/firebase/functions/GeneratePoemFunction.java
index 329f7c81c..5ae8cac05 100644
--- a/samples/firebase/src/main/java/com/google/genkit/samples/firebase/functions/GeneratePoemFunction.java
+++ b/samples/firebase/src/main/java/com/google/genkit/samples/firebase/functions/GeneratePoemFunction.java
@@ -69,7 +69,7 @@ public GeneratePoemFunction() {
           ModelResponse response =
               genkit.generate(
                   GenerateOptions.builder()
-                      .model("googleai/gemini-2.0-flash")
+                      .model("googleai/gemini-2.5-flash")
                       .prompt("Write a short, creative poem about: " + topic)
                       .config(
                           GenerationConfig.builder().temperature(0.9).maxOutputTokens(500).build())
diff --git a/samples/google-genai/README.md b/samples/google-genai/README.md
index 6f8be0c4a..949330d1c 100644
--- a/samples/google-genai/README.md
+++ b/samples/google-genai/README.md
@@ -118,11 +118,11 @@ The Google GenAI plugin provides access to:
 
 | Model | Description |
 |-------|-------------|
-| `googleai/gemini-2.0-flash` | Fast, efficient Gemini model |
+| `googleai/gemini-2.5-flash` | Fast, efficient Gemini model |
 | `googleai/gemini-1.5-pro` | Advanced reasoning capabilities |
 | `googleai/gemini-1.5-flash` | Balanced speed and capability |
 | `googleai/imagen-3.0-generate-002` | Image generation |
-| `googleai/text-embedding-004` | Text embeddings |
+| `googleai/gemini-embedding-001` | Text embeddings |
 
 ## Code Highlights
 
@@ -148,7 +148,7 @@ genkit.defineFlow("textGeneration", String.class, String.class,
     (ctx, prompt) -> {
         ModelResponse response = genkit.generate(
             GenerateOptions.builder()
-                .model("googleai/gemini-2.0-flash")
+                .model("googleai/gemini-2.5-flash")
                 .prompt(prompt)
                 .config(GenerationConfig.builder()
                     .temperature(0.7)
diff --git a/samples/google-genai/src/main/java/com/google/genkit/samples/GoogleGenAIApp.java b/samples/google-genai/src/main/java/com/google/genkit/samples/GoogleGenAIApp.java
index 6cb22b403..4558b5d86 100644
--- a/samples/google-genai/src/main/java/com/google/genkit/samples/GoogleGenAIApp.java
+++ b/samples/google-genai/src/main/java/com/google/genkit/samples/GoogleGenAIApp.java
@@ -145,7 +145,7 @@ private static void defineTextGenerationFlow(Genkit genkit) {
           ModelResponse response =
               genkit.generate(
                   GenerateOptions.builder()
-                      .model("googleai/gemini-2.0-flash")
+                      .model("googleai/gemini-2.5-flash")
                       .prompt(prompt)
                       .config(config)
                       .build());
@@ -201,7 +201,7 @@ private static void defineToolCallingFlow(Genkit genkit) {
           ModelResponse response =
               genkit.generate(
                   GenerateOptions.builder()
-                      .model("googleai/gemini-2.0-flash")
+                      .model("googleai/gemini-2.5-flash")
                       .prompt(prompt)
                       .tools(List.of(weatherTool))
                       .build());
@@ -217,7 +217,7 @@ private static void defineEmbeddingsFlow(Genkit genkit) {
         String.class,
         (ctx, text) -> {
           List documents = Arrays.asList(Document.fromText(text));
-          EmbedResponse response = genkit.embed("googleai/text-embedding-004", documents);
+          EmbedResponse response = genkit.embed("googleai/gemini-embedding-001", documents);
 
           if (response.getEmbeddings() != null && !response.getEmbeddings().isEmpty()) {
             EmbedResponse.Embedding embedding = response.getEmbeddings().get(0);
diff --git a/samples/mongo-vector/README.md b/samples/mongo-vector/README.md
new file mode 100644
index 000000000..6f8461088
--- /dev/null
+++ b/samples/mongo-vector/README.md
@@ -0,0 +1,65 @@
+# MongoDB Vector RAG Sample
+
+A Retrieval-Augmented Generation sample that indexes film descriptions into **MongoDB Atlas Vector Search** via `MongoPlugin` and answers questions with a **Gemini** model.
+
+## Prerequisites
+
+- Java 21+ and Maven 3.6+
+- A `GEMINI_API_KEY` — get one from [Google AI Studio](https://aistudio.google.com/apikey)
+- A MongoDB deployment that supports Atlas Vector Search — a MongoDB Atlas cluster or the local Atlas image (see below). A plain `mongo` server does **not** support `$vectorSearch`.
+
+## Run MongoDB Atlas locally with Docker
+
+Use the official Atlas Local image, which bundles the search/vector engine:
+
+```bash
+docker run --detach \
+  --name genkit-atlas \
+  --publish 27017:27017 \
+  mongodb/mongodb-atlas-local:latest
+
+# wait until it's accepting connections
+until docker exec genkit-atlas mongosh --quiet --eval 'db.runCommand({ ping: 1 })' >/dev/null 2>&1; do sleep 1; done
+echo "atlas-local ready"
+```
+
+The database (`genkit`), collection (`films`), and the Atlas Vector Search index are created automatically on first use. The index takes a few seconds to become queryable. Stop and remove the container later with:
+
+```bash
+docker rm -f genkit-atlas
+```
+
+## Configure
+
+```bash
+export GEMINI_API_KEY=
+export MONGO_URI="mongodb://localhost:27017/?directConnection=true"   # optional (default)
+export MONGO_DATABASE=genkit                                          # optional (default)
+export MONGO_COLLECTION=films                                         # optional (default)
+```
+
+For a MongoDB Atlas cluster, set `MONGO_URI` to your `mongodb+srv://...` connection string.
+
+## Run
+
+```bash
+mvn -q exec:java
+```
+
+Then open the Genkit Dev UI at http://localhost:4000 (or run under `genkit start -- mvn -q exec:java`) and exercise the flows:
+
+- `indexDocuments` — index the sample film descriptions (also creates the vector index on first run)
+- `retrieveDocuments` — return the films matching a query
+- `ragQuery` — answer a question using retrieved context
+
+Or via curl:
+
+```bash
+curl -X POST http://localhost:8080/api/flows/indexDocuments -H 'Content-Type: application/json' -d '{}'
+curl -X POST http://localhost:8080/api/flows/ragQuery -H 'Content-Type: application/json' \
+  -d '{"data": "What Christopher Nolan films are mentioned?"}'
+```
+
+## Configuration
+
+`MongoPlugin` uses the `$vectorSearch` aggregation stage. Tune per-collection settings with `MongoVectorStoreConfig` (database/collection names, embedder, index name, dimension, similarity, text/embedding field names, `numCandidates`, `createIndexIfNotExists`).
diff --git a/samples/mongo-vector/pom.xml b/samples/mongo-vector/pom.xml
new file mode 100644
index 000000000..0bb281c67
--- /dev/null
+++ b/samples/mongo-vector/pom.xml
@@ -0,0 +1,86 @@
+
+
+
+    4.0.0
+
+    
+        com.google.genkit
+        genkit-parent
+        1.0.0-SNAPSHOT
+        ../../pom.xml
+    
+
+    genkit-sample-mongo-vector
+    jar
+    Genkit MongoDB Vector RAG Sample
+    Sample application demonstrating MongoDB Atlas Vector Search with Genkit
+
+    
+        true
+        com.google.genkit.samples.mongo.MongoVectorRAGSample
+    
+
+    
+        
+            com.google.genkit
+            genkit
+            ${project.version}
+        
+        
+            com.google.genkit
+            genkit-plugin-mongodb
+            ${project.version}
+        
+        
+            com.google.genkit
+            genkit-plugin-google-genai
+            ${project.version}
+        
+        
+            com.google.genkit
+            genkit-plugin-jetty
+            ${project.version}
+        
+        
+            ch.qos.logback
+            logback-classic
+        
+        
+            io.github.cdimascio
+            dotenv-java
+            3.2.0
+        
+    
+
+    
+        
+            
+                org.codehaus.mojo
+                exec-maven-plugin
+                3.6.3
+                
+                    ${exec.mainClass}
+                
+            
+        
+    
+
diff --git a/samples/mongo-vector/run.sh b/samples/mongo-vector/run.sh
new file mode 100755
index 000000000..d10fb3e2a
--- /dev/null
+++ b/samples/mongo-vector/run.sh
@@ -0,0 +1,7 @@
+#!/bin/bash
+
+# Genkit Sample Runner
+# This script runs the sample application
+
+echo "Building and running the sample..."
+mvn clean compile exec:java
diff --git a/samples/mongo-vector/src/main/java/com/google/genkit/samples/mongo/MongoVectorRAGSample.java b/samples/mongo-vector/src/main/java/com/google/genkit/samples/mongo/MongoVectorRAGSample.java
new file mode 100644
index 000000000..16ed36d2c
--- /dev/null
+++ b/samples/mongo-vector/src/main/java/com/google/genkit/samples/mongo/MongoVectorRAGSample.java
@@ -0,0 +1,171 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package com.google.genkit.samples.mongo;
+
+import com.google.genkit.Genkit;
+import com.google.genkit.GenkitOptions;
+import com.google.genkit.ai.*;
+import com.google.genkit.core.Flow;
+import com.google.genkit.plugins.googlegenai.GoogleGenAIPlugin;
+import com.google.genkit.plugins.jetty.JettyPlugin;
+import com.google.genkit.plugins.jetty.JettyPluginOptions;
+import com.google.genkit.plugins.mongodb.MongoPlugin;
+import com.google.genkit.plugins.mongodb.MongoVectorStoreConfig;
+import io.github.cdimascio.dotenv.Dotenv;
+import java.util.List;
+import java.util.stream.Collectors;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+/**
+ * Sample application demonstrating MongoDB Atlas Vector Search with Genkit.
+ *
+ * 
Indexes a handful of film descriptions into a MongoDB collection, retrieves the nearest
+ * matches for a query with the {@code $vectorSearch} stage, and answers questions with a RAG flow.
+ * Requires a MongoDB Atlas cluster or the {@code mongodb/mongodb-atlas-local} Docker image (see the
+ * README) and a {@code GEMINI_API_KEY}.
+ */
+public class MongoVectorRAGSample {
+
+  private static final Logger logger = LoggerFactory.getLogger(MongoVectorRAGSample.class);
+
+  private static final List SAMPLE_DOCUMENTS =
+      List.of(
+          "The Godfather is a 1972 crime film directed by Francis Ford Coppola about the Corleone crime family.",
+          "The Dark Knight is a 2008 superhero film directed by Christopher Nolan featuring Batman against the Joker.",
+          "Pulp Fiction is a 1994 crime film directed by Quentin Tarantino known for its nonlinear narrative.",
+          "Inception is a 2010 sci-fi film directed by Christopher Nolan about dream infiltration.",
+          "The Matrix is a 1999 sci-fi film directed by the Wachowskis exploring simulated reality.",
+          "Forrest Gump is a 1994 drama directed by Robert Zemeckis about a man's extraordinary life.",
+          "Star Wars is a 1977 sci-fi film directed by George Lucas set in a galaxy far, far away.",
+          "The Shawshank Redemption is a 1994 drama about hope and friendship in a prison.");
+
+  private static final String RAG_SYSTEM_PROMPT =
+      """
+      You are a helpful assistant that answers questions based on the provided context documents.
+      Answer only from the context. If the context is insufficient, say so.
+      """;
+
+  public static void main(String[] args) {
+    Dotenv dotenv = Dotenv.configure().ignoreIfMissing().systemProperties().load();
+
+    String geminiApiKey = getEnv(dotenv, "GEMINI_API_KEY");
+    if (geminiApiKey == null) {
+      logger.error("Please set GEMINI_API_KEY in .env file or environment variable");
+      System.exit(1);
+    }
+
+    String mongoUri =
+        getEnvOrDefault(dotenv, "MONGO_URI", "mongodb://localhost:27017/?directConnection=true");
+    String database = getEnvOrDefault(dotenv, "MONGO_DATABASE", "genkit");
+    String collection = getEnvOrDefault(dotenv, "MONGO_COLLECTION", "films");
+
+    logger.info("Starting MongoDB Vector RAG Sample (db={}, collection={})", database, collection);
+
+    MongoPlugin mongoPlugin =
+        MongoPlugin.builder()
+            .connectionString(mongoUri)
+            .addCollection(
+                MongoVectorStoreConfig.builder()
+                    .databaseName(database)
+                    .collectionName(collection)
+                    .embedderName("googleai/gemini-embedding-001")
+                    .dimension(768)
+                    .similarity(MongoVectorStoreConfig.Similarity.COSINE)
+                    .createIndexIfNotExists(true)
+                    .build())
+            .build();
+
+    JettyPlugin jetty = new JettyPlugin(JettyPluginOptions.builder().port(8080).build());
+    Genkit genkit =
+        Genkit.builder()
+            .options(GenkitOptions.builder().devMode(true).reflectionPort(3100).build())
+            .plugin(GoogleGenAIPlugin.create(geminiApiKey))
+            .plugin(mongoPlugin)
+            .plugin(jetty)
+            .build();
+
+    String action = "mongodb/" + collection;
+
+    Flow indexDocumentsFlow =
+        genkit.defineFlow(
+            "indexDocuments",
+            Void.class,
+            String.class,
+            (ctx, input) -> {
+              List documents =
+                  SAMPLE_DOCUMENTS.stream().map(Document::fromText).collect(Collectors.toList());
+              genkit.index(action, documents);
+              return "Successfully indexed " + documents.size() + " documents";
+            });
+
+    @SuppressWarnings("unchecked")
+    Flow, Void> retrieveDocumentsFlow =
+        genkit.defineFlow(
+            "retrieveDocuments",
+            String.class,
+            (Class>) (Class) List.class,
+            (ctx, query) -> {
+              List docs = genkit.retrieve(action, query);
+              return docs.stream().map(Document::text).collect(Collectors.toList());
+            });
+
+    Flow ragQueryFlow =
+        genkit.defineFlow(
+            "ragQuery",
+            String.class,
+            String.class,
+            (ctx, question) -> {
+              List docs = genkit.retrieve(action, question);
+              ModelResponse response =
+                  genkit.generate(
+                      GenerateOptions.builder()
+                          .model("googleai/gemini-2.5-flash")
+                          .system(RAG_SYSTEM_PROMPT)
+                          .prompt(question)
+                          .docs(docs)
+                          .config(GenerationConfig.builder().temperature(0.3).build())
+                          .build());
+              return response.getText();
+            });
+
+    logger.info(
+        "Genkit MongoDB Vector RAG Sample started. Flows: indexDocuments, retrieveDocuments, ragQuery");
+    logger.info("Dev UI: http://localhost:4000  |  Reflection: http://localhost:3100");
+    logger.info(
+        "Note: the Atlas Vector Search index is created on first index/retrieve and may take a moment to become queryable.");
+
+    try {
+      jetty.start();
+    } catch (Exception e) {
+      logger.error("Failed to start Jetty server", e);
+      System.exit(1);
+    }
+  }
+
+  private static String getEnv(Dotenv dotenv, String name) {
+    String value = dotenv.get(name);
+    return (value != null && !value.isBlank()) ? value : System.getenv(name);
+  }
+
+  private static String getEnvOrDefault(Dotenv dotenv, String name, String defaultValue) {
+    String value = getEnv(dotenv, name);
+    return (value != null && !value.isBlank()) ? value : defaultValue;
+  }
+}
diff --git a/samples/mongo-vector/src/main/resources/logback.xml b/samples/mongo-vector/src/main/resources/logback.xml
new file mode 100644
index 000000000..47172d8f9
--- /dev/null
+++ b/samples/mongo-vector/src/main/resources/logback.xml
@@ -0,0 +1,21 @@
+
+
+    
+        
+            %d{HH:mm:ss.SSS} %-5level %logger{24} - %msg%n
+        
+    
+
+    
+    
+    
+    
+    
+    
+
+    
+
+    
+        
+    
+
diff --git a/samples/pinecone/README.md b/samples/pinecone/README.md
index da8faa3c1..4dfdcca16 100644
--- a/samples/pinecone/README.md
+++ b/samples/pinecone/README.md
@@ -131,8 +131,8 @@ curl -X POST http://localhost:4000/api/flows/ragQuery \
 ## Configuration
 
 The sample uses:
-- **Embedder**: `googleai/text-embedding-004` (768 dimensions)
-- **LLM**: `googleai/gemini-2.0-flash`
+- **Embedder**: `googleai/gemini-embedding-001` (768 dimensions)
+- **LLM**: `googleai/gemini-2.5-flash`
 - **Metric**: Cosine similarity
 - **Index Type**: Serverless (AWS us-east-1)
 
@@ -146,7 +146,7 @@ Pinecone supports namespaces for multi-tenant applications. To use namespaces:
 .addIndex(PineconeIndexConfig.builder()
     .indexName("my-index")
     .namespace("production")  // Add namespace
-    .embedderName("googleai/text-embedding-004")
+    .embedderName("googleai/gemini-embedding-001")
     .build())
 ```
 
diff --git a/samples/pinecone/src/main/java/com/google/genkit/samples/pinecone/PineconeRAGSample.java b/samples/pinecone/src/main/java/com/google/genkit/samples/pinecone/PineconeRAGSample.java
index 45bafa9cf..b9a768f02 100644
--- a/samples/pinecone/src/main/java/com/google/genkit/samples/pinecone/PineconeRAGSample.java
+++ b/samples/pinecone/src/main/java/com/google/genkit/samples/pinecone/PineconeRAGSample.java
@@ -118,7 +118,7 @@ public static void main(String[] args) {
             .addIndex(
                 PineconeIndexConfig.builder()
                     .indexName(finalIndexName)
-                    .embedderName("googleai/text-embedding-004")
+                    .embedderName("googleai/gemini-embedding-001")
                     .dimension(768)
                     .metric(PineconeIndexConfig.Metric.COSINE)
                     .cloud(cloudEnum)
@@ -199,7 +199,7 @@ public static void main(String[] args) {
               ModelResponse response =
                   genkit.generate(
                       GenerateOptions.builder()
-                          .model("googleai/gemini-2.0-flash")
+                          .model("googleai/gemini-2.5-flash")
                           .system(RAG_SYSTEM_PROMPT)
                           .prompt(question)
                           .docs(docs)
diff --git a/samples/postgresql/README.md b/samples/postgresql/README.md
index a593ad4d2..f5339058a 100644
--- a/samples/postgresql/README.md
+++ b/samples/postgresql/README.md
@@ -140,8 +140,8 @@ WITH (lists = 100);
 ## Configuration
 
 The sample uses:
-- **Embedder**: `googleai/text-embedding-004` (768 dimensions)
-- **LLM**: `googleai/gemini-2.0-flash`
+- **Embedder**: `googleai/gemini-embedding-001` (768 dimensions)
+- **LLM**: `googleai/gemini-2.5-flash`
 - **Distance Strategy**: Cosine distance
 - **Table Name**: `films`
 - **Index Type**: IVFFlat with 100 lists
diff --git a/samples/postgresql/src/main/java/com/google/genkit/samples/postgresql/PostgresRAGSample.java b/samples/postgresql/src/main/java/com/google/genkit/samples/postgresql/PostgresRAGSample.java
index 8d562f79c..f023f1013 100644
--- a/samples/postgresql/src/main/java/com/google/genkit/samples/postgresql/PostgresRAGSample.java
+++ b/samples/postgresql/src/main/java/com/google/genkit/samples/postgresql/PostgresRAGSample.java
@@ -116,7 +116,7 @@ public static void main(String[] args) {
             .addTable(
                 PostgresTableConfig.builder()
                     .tableName("films")
-                    .embedderName("googleai/text-embedding-004")
+                    .embedderName("googleai/gemini-embedding-001")
                     .vectorDimension(768)
                     .distanceStrategy(PostgresTableConfig.DistanceStrategy.L2)
                     .createTableIfNotExists(true)
@@ -194,7 +194,7 @@ public static void main(String[] args) {
               ModelResponse response =
                   genkit.generate(
                       GenerateOptions.builder()
-                          .model("googleai/gemini-2.0-flash")
+                          .model("googleai/gemini-2.5-flash")
                           .system(RAG_SYSTEM_PROMPT)
                           .prompt(question)
                           .docs(docs)
diff --git a/samples/qdrant/README.md b/samples/qdrant/README.md
new file mode 100644
index 000000000..9b0423c0d
--- /dev/null
+++ b/samples/qdrant/README.md
@@ -0,0 +1,63 @@
+# Qdrant RAG Sample
+
+A Retrieval-Augmented Generation sample that indexes film descriptions into **Qdrant** via `QdrantPlugin` and answers questions with a **Gemini** model.
+
+## Prerequisites
+
+- Java 21+ and Maven 3.6+
+- A `GEMINI_API_KEY` — get one from [Google AI Studio](https://aistudio.google.com/apikey)
+- A reachable Qdrant server (see the Docker command below)
+
+## Run Qdrant locally with Docker
+
+```bash
+docker run --detach \
+  --name genkit-qdrant \
+  --publish 6333:6333 \
+  qdrant/qdrant
+
+# wait until it's accepting connections
+until curl -sf http://localhost:6333/readyz >/dev/null 2>&1; do sleep 1; done
+echo "qdrant ready"
+```
+
+The collection is created automatically on first use (dimension 768, cosine distance). Stop and remove the container later with:
+
+```bash
+docker rm -f genkit-qdrant
+```
+
+## Configure
+
+```bash
+export GEMINI_API_KEY=
+export QDRANT_URL=http://localhost:6333        # optional (default)
+export QDRANT_API_KEY=               # optional (required for Qdrant Cloud)
+export QDRANT_COLLECTION=genkit_films          # optional (default)
+```
+
+## Run
+
+```bash
+mvn -q exec:java
+```
+
+Then open the Genkit Dev UI at http://localhost:4000 (or run under `genkit start -- mvn -q exec:java`) and exercise the flows:
+
+- `indexDocuments` — index the sample film descriptions
+- `retrieveDocuments` — return the films matching a query
+- `ragQuery` — answer a question using retrieved context
+
+Or via curl:
+
+```bash
+curl -X POST http://localhost:4000/api/flows/indexDocuments -H 'Content-Type: application/json' -d '{}'
+curl -X POST http://localhost:4000/api/flows/ragQuery -H 'Content-Type: application/json' \
+  -d '{"data": "What Christopher Nolan films are mentioned?"}'
+```
+
+Inspect stored points in the Qdrant dashboard at http://localhost:6333/dashboard.
+
+## Configuration
+
+`QdrantPlugin` talks to the Qdrant REST API. Tune per-collection settings with `QdrantCollectionConfig` (collection name, embedder, dimension, distance function, text payload key, `createCollectionIfNotExists`, additional metadata).
diff --git a/samples/qdrant/pom.xml b/samples/qdrant/pom.xml
new file mode 100644
index 000000000..3ca14a4cf
--- /dev/null
+++ b/samples/qdrant/pom.xml
@@ -0,0 +1,86 @@
+
+
+
+    4.0.0
+
+    
+        com.google.genkit
+        genkit-parent
+        1.0.0-SNAPSHOT
+        ../../pom.xml
+    
+
+    genkit-sample-qdrant
+    jar
+    Genkit Qdrant RAG Sample
+    Sample application demonstrating Qdrant vector search with Genkit
+
+    
+        true
+        com.google.genkit.samples.qdrant.QdrantRAGSample
+    
+
+    
+        
+            com.google.genkit
+            genkit
+            ${project.version}
+        
+        
+            com.google.genkit
+            genkit-plugin-qdrant
+            ${project.version}
+        
+        
+            com.google.genkit
+            genkit-plugin-google-genai
+            ${project.version}
+        
+        
+            com.google.genkit
+            genkit-plugin-jetty
+            ${project.version}
+        
+        
+            ch.qos.logback
+            logback-classic
+        
+        
+            io.github.cdimascio
+            dotenv-java
+            3.2.0
+        
+    
+
+    
+        
+            
+                org.codehaus.mojo
+                exec-maven-plugin
+                3.6.3
+                
+                    ${exec.mainClass}
+                
+            
+        
+    
+
diff --git a/samples/qdrant/run.sh b/samples/qdrant/run.sh
new file mode 100755
index 000000000..d10fb3e2a
--- /dev/null
+++ b/samples/qdrant/run.sh
@@ -0,0 +1,7 @@
+#!/bin/bash
+
+# Genkit Sample Runner
+# This script runs the sample application
+
+echo "Building and running the sample..."
+mvn clean compile exec:java
diff --git a/samples/qdrant/src/main/java/com/google/genkit/samples/qdrant/QdrantRAGSample.java b/samples/qdrant/src/main/java/com/google/genkit/samples/qdrant/QdrantRAGSample.java
new file mode 100644
index 000000000..396a131b3
--- /dev/null
+++ b/samples/qdrant/src/main/java/com/google/genkit/samples/qdrant/QdrantRAGSample.java
@@ -0,0 +1,166 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package com.google.genkit.samples.qdrant;
+
+import com.google.genkit.Genkit;
+import com.google.genkit.GenkitOptions;
+import com.google.genkit.ai.*;
+import com.google.genkit.core.Flow;
+import com.google.genkit.plugins.googlegenai.GoogleGenAIPlugin;
+import com.google.genkit.plugins.jetty.JettyPlugin;
+import com.google.genkit.plugins.jetty.JettyPluginOptions;
+import com.google.genkit.plugins.qdrant.QdrantCollectionConfig;
+import com.google.genkit.plugins.qdrant.QdrantPlugin;
+import io.github.cdimascio.dotenv.Dotenv;
+import java.util.List;
+import java.util.stream.Collectors;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+/**
+ * Sample application demonstrating Qdrant vector search with Genkit.
+ *
+ * 
Indexes a handful of film descriptions into a Qdrant collection, retrieves the nearest matches
+ * for a query, and answers questions with a RAG flow. Requires a running Qdrant server (see the
+ * README for a one-line Docker command) and a {@code GEMINI_API_KEY}.
+ */
+public class QdrantRAGSample {
+
+  private static final Logger logger = LoggerFactory.getLogger(QdrantRAGSample.class);
+
+  private static final List SAMPLE_DOCUMENTS =
+      List.of(
+          "The Godfather is a 1972 crime film directed by Francis Ford Coppola about the Corleone crime family.",
+          "The Dark Knight is a 2008 superhero film directed by Christopher Nolan featuring Batman against the Joker.",
+          "Pulp Fiction is a 1994 crime film directed by Quentin Tarantino known for its nonlinear narrative.",
+          "Inception is a 2010 sci-fi film directed by Christopher Nolan about dream infiltration.",
+          "The Matrix is a 1999 sci-fi film directed by the Wachowskis exploring simulated reality.",
+          "Forrest Gump is a 1994 drama directed by Robert Zemeckis about a man's extraordinary life.",
+          "Star Wars is a 1977 sci-fi film directed by George Lucas set in a galaxy far, far away.",
+          "The Shawshank Redemption is a 1994 drama about hope and friendship in a prison.");
+
+  private static final String RAG_SYSTEM_PROMPT =
+      """
+      You are a helpful assistant that answers questions based on the provided context documents.
+      Answer only from the context. If the context is insufficient, say so.
+      """;
+
+  public static void main(String[] args) {
+    Dotenv dotenv = Dotenv.configure().ignoreIfMissing().systemProperties().load();
+
+    String geminiApiKey = getEnv(dotenv, "GEMINI_API_KEY");
+    if (geminiApiKey == null) {
+      logger.error("Please set GEMINI_API_KEY in .env file or environment variable");
+      System.exit(1);
+    }
+
+    String qdrantUrl = getEnvOrDefault(dotenv, "QDRANT_URL", "http://localhost:6333");
+    String qdrantApiKey = getEnv(dotenv, "QDRANT_API_KEY"); // optional
+    String collection = getEnvOrDefault(dotenv, "QDRANT_COLLECTION", "genkit_films");
+
+    logger.info("Starting Qdrant RAG Sample (url={}, collection={})", qdrantUrl, collection);
+
+    QdrantPlugin qdrantPlugin =
+        QdrantPlugin.builder()
+            .url(qdrantUrl)
+            .apiKey(qdrantApiKey)
+            .addCollection(
+                QdrantCollectionConfig.builder()
+                    .collectionName(collection)
+                    .embedderName("googleai/gemini-embedding-001")
+                    .dimension(768)
+                    .distance(QdrantCollectionConfig.Distance.COSINE)
+                    .build())
+            .build();
+
+    JettyPlugin jetty = new JettyPlugin(JettyPluginOptions.builder().port(8080).build());
+    Genkit genkit =
+        Genkit.builder()
+            .options(GenkitOptions.builder().devMode(true).reflectionPort(3100).build())
+            .plugin(GoogleGenAIPlugin.create(geminiApiKey))
+            .plugin(qdrantPlugin)
+            .plugin(jetty)
+            .build();
+
+    String action = "qdrant/" + collection;
+
+    Flow indexDocumentsFlow =
+        genkit.defineFlow(
+            "indexDocuments",
+            Void.class,
+            String.class,
+            (ctx, input) -> {
+              List documents =
+                  SAMPLE_DOCUMENTS.stream().map(Document::fromText).collect(Collectors.toList());
+              genkit.index(action, documents);
+              return "Successfully indexed " + documents.size() + " documents";
+            });
+
+    @SuppressWarnings("unchecked")
+    Flow, Void> retrieveDocumentsFlow =
+        genkit.defineFlow(
+            "retrieveDocuments",
+            String.class,
+            (Class>) (Class) List.class,
+            (ctx, query) -> {
+              List docs = genkit.retrieve(action, query);
+              return docs.stream().map(Document::text).collect(Collectors.toList());
+            });
+
+    Flow ragQueryFlow =
+        genkit.defineFlow(
+            "ragQuery",
+            String.class,
+            String.class,
+            (ctx, question) -> {
+              List docs = genkit.retrieve(action, question);
+              ModelResponse response =
+                  genkit.generate(
+                      GenerateOptions.builder()
+                          .model("googleai/gemini-2.5-flash")
+                          .system(RAG_SYSTEM_PROMPT)
+                          .prompt(question)
+                          .docs(docs)
+                          .config(GenerationConfig.builder().temperature(0.3).build())
+                          .build());
+              return response.getText();
+            });
+
+    logger.info(
+        "Genkit Qdrant RAG Sample started. Flows: indexDocuments, retrieveDocuments, ragQuery");
+    logger.info("Dev UI: http://localhost:4000  |  Reflection: http://localhost:3100");
+
+    try {
+      jetty.start();
+    } catch (Exception e) {
+      logger.error("Failed to start Jetty server", e);
+      System.exit(1);
+    }
+  }
+
+  private static String getEnv(Dotenv dotenv, String name) {
+    String value = dotenv.get(name);
+    return (value != null && !value.isBlank()) ? value : System.getenv(name);
+  }
+
+  private static String getEnvOrDefault(Dotenv dotenv, String name, String defaultValue) {
+    String value = getEnv(dotenv, name);
+    return (value != null && !value.isBlank()) ? value : defaultValue;
+  }
+}
diff --git a/samples/qdrant/src/main/resources/logback.xml b/samples/qdrant/src/main/resources/logback.xml
new file mode 100644
index 000000000..dff2540b6
--- /dev/null
+++ b/samples/qdrant/src/main/resources/logback.xml
@@ -0,0 +1,20 @@
+
+
+    
+        
+            %d{HH:mm:ss.SSS} %-5level %logger{24} - %msg%n
+        
+    
+
+    
+    
+    
+    
+    
+
+    
+
+    
+        
+    
+
diff --git a/samples/weaviate/README.md b/samples/weaviate/README.md
index c662a095b..676547a84 100644
--- a/samples/weaviate/README.md
+++ b/samples/weaviate/README.md
@@ -119,8 +119,8 @@ curl -X POST http://localhost:4000/api/flows/ragQuery \
 ## Configuration
 
 The sample uses:
-- **Embedder**: `googleai/text-embedding-004` (768 dimensions)
-- **LLM**: `googleai/gemini-2.0-flash`
+- **Embedder**: `googleai/gemini-embedding-001` (768 dimensions)
+- **LLM**: `googleai/gemini-2.5-flash`
 - **Distance Metric**: Cosine similarity
 - **Collection Name**: `Films`
 
diff --git a/samples/weaviate/src/main/java/com/google/genkit/samples/weaviate/WeaviateRAGSample.java b/samples/weaviate/src/main/java/com/google/genkit/samples/weaviate/WeaviateRAGSample.java
index 6c1a89ac7..643af8c59 100644
--- a/samples/weaviate/src/main/java/com/google/genkit/samples/weaviate/WeaviateRAGSample.java
+++ b/samples/weaviate/src/main/java/com/google/genkit/samples/weaviate/WeaviateRAGSample.java
@@ -117,7 +117,7 @@ public static void main(String[] args) {
     weaviateBuilder.addCollection(
         WeaviateCollectionConfig.builder()
             .name("Films")
-            .embedderName("googleai/text-embedding-004")
+            .embedderName("googleai/gemini-embedding-001")
             .vectorDimension(768)
             .distanceMeasure(WeaviateCollectionConfig.DistanceMeasure.COSINE)
             .createCollectionIfMissing(true)
@@ -192,7 +192,7 @@ public static void main(String[] args) {
               ModelResponse response =
                   genkit.generate(
                       GenerateOptions.builder()
-                          .model("googleai/gemini-2.0-flash")
+                          .model("googleai/gemini-2.5-flash")
                           .system(RAG_SYSTEM_PROMPT)
                           .prompt(question)
                           .docs(docs)
diff --git a/skills/building-ai-apps-with-genkit-java/SKILL.md b/skills/building-ai-apps-with-genkit-java/SKILL.md
index a816471b9..637a7edf8 100644
--- a/skills/building-ai-apps-with-genkit-java/SKILL.md
+++ b/skills/building-ai-apps-with-genkit-java/SKILL.md
@@ -169,8 +169,8 @@ openai/text-embedding-3-small, openai/text-embedding-3-large
 openai/dall-e-3, openai/dall-e-2, openai/gpt-image-1
 
 # Google Gemini
-googleai/gemini-2.0-flash, googleai/gemini-1.5-pro, googleai/gemini-1.5-flash
-googleai/text-embedding-004, googleai/imagen-3.0-generate-002
+googleai/gemini-2.5-flash, googleai/gemini-1.5-pro, googleai/gemini-1.5-flash
+googleai/gemini-embedding-001, googleai/imagen-3.0-generate-002
 
 # Anthropic
 anthropic/claude-sonnet-4-5-20250929, anthropic/claude-opus-4-5-20251101
@@ -801,7 +801,7 @@ Genkit genkit = Genkit.builder()
         .addRetriever(FirestoreRetrieverConfig.builder()
             .name("my-docs")
             .collection("documents")
-            .embedderName("googleai/text-embedding-004")
+            .embedderName("googleai/gemini-embedding-001")
             .vectorField("embedding")
             .contentField("text")
             .build())
@@ -827,7 +827,7 @@ public class MyFunction implements HttpFunction {
         genkit.defineFlow("myFlow", String.class, String.class,
             (ctx, input) -> genkit.generate(
                 GenerateOptions.builder()
-                    .model("googleai/gemini-2.0-flash")
+                    .model("googleai/gemini-2.5-flash")
                     .prompt(input)
                     .build()).getText());
 
@@ -954,7 +954,7 @@ Genkit is provider-agnostic. Change the model string and plugin:
 // Switch from OpenAI to Gemini — just change plugin + model name
 .plugin(GoogleGenAIPlugin.create())
 // ...
-.model("googleai/gemini-2.0-flash")
+.model("googleai/gemini-2.5-flash")
 
 // Switch to Anthropic
 .plugin(AnthropicPlugin.create())
@@ -987,7 +987,7 @@ genkit.defineFlow("deepAnalysis", String.class, String.class,
 
 genkit.defineFlow("creative", String.class, String.class,
     (ctx, q) -> genkit.generate(GenerateOptions.builder()
-        .model("googleai/gemini-2.0-flash").prompt(q).build()).getText());
+        .model("googleai/gemini-2.5-flash").prompt(q).build()).getText());
 ```
 
 ### Custom OpenAI-Compatible Endpoint

From 65a72774fdb006883b8550f5c69bcbb539e5139c Mon Sep 17 00:00:00 2001
From: xavidop 
Date: Mon, 6 Jul 2026 14:34:57 +0200
Subject: [PATCH 2/6] feat: milvus plugin

---
 .gitignore                                    |   4 +-
 docs/astro.config.mjs                         |   1 +
 docs/src/content/docs/index.mdx               |   1 +
 docs/src/content/docs/plugins/milvus.md       |  72 ++++
 plugins/milvus/pom.xml                        |  78 +++++
 .../milvus/MilvusCollectionConfig.java        | 235 +++++++++++++
 .../genkit/plugins/milvus/MilvusPlugin.java   | 175 ++++++++++
 .../plugins/milvus/MilvusVectorStore.java     | 321 ++++++++++++++++++
 .../genkit/plugins/milvus/package-info.java   |  25 ++
 .../milvus/MilvusCollectionConfigTest.java    |  75 ++++
 .../plugins/milvus/MilvusPluginTest.java      |  48 +++
 .../plugins/milvus/MilvusVectorStoreTest.java | 118 +++++++
 pom.xml                                       |   2 +
 samples/README.md                             |   1 +
 samples/milvus/README.md                      |  61 ++++
 samples/milvus/pom.xml                        |  86 +++++
 samples/milvus/run.sh                         |   7 +
 .../samples/milvus/MilvusRAGSample.java       | 165 +++++++++
 samples/milvus/src/main/resources/logback.xml |  20 ++
 samples/milvus/user.yaml                      |   1 +
 20 files changed, 1495 insertions(+), 1 deletion(-)
 create mode 100644 docs/src/content/docs/plugins/milvus.md
 create mode 100644 plugins/milvus/pom.xml
 create mode 100644 plugins/milvus/src/main/java/com/google/genkit/plugins/milvus/MilvusCollectionConfig.java
 create mode 100644 plugins/milvus/src/main/java/com/google/genkit/plugins/milvus/MilvusPlugin.java
 create mode 100644 plugins/milvus/src/main/java/com/google/genkit/plugins/milvus/MilvusVectorStore.java
 create mode 100644 plugins/milvus/src/main/java/com/google/genkit/plugins/milvus/package-info.java
 create mode 100644 plugins/milvus/src/test/java/com/google/genkit/plugins/milvus/MilvusCollectionConfigTest.java
 create mode 100644 plugins/milvus/src/test/java/com/google/genkit/plugins/milvus/MilvusPluginTest.java
 create mode 100644 plugins/milvus/src/test/java/com/google/genkit/plugins/milvus/MilvusVectorStoreTest.java
 create mode 100644 samples/milvus/README.md
 create mode 100644 samples/milvus/pom.xml
 create mode 100755 samples/milvus/run.sh
 create mode 100644 samples/milvus/src/main/java/com/google/genkit/samples/milvus/MilvusRAGSample.java
 create mode 100644 samples/milvus/src/main/resources/logback.xml
 create mode 100644 samples/milvus/user.yaml

diff --git a/.gitignore b/.gitignore
index 13ca98480..eb627cbf3 100644
--- a/.gitignore
+++ b/.gitignore
@@ -31,4 +31,6 @@ node_modules
 hs_err_pid*
 samples/google-genai/generated_media/
 .astro
-.snapshots
\ No newline at end of file
+.snapshots
+
+samples/milvus/volumes
\ No newline at end of file
diff --git a/docs/astro.config.mjs b/docs/astro.config.mjs
index 9bae49eff..8f511185a 100644
--- a/docs/astro.config.mjs
+++ b/docs/astro.config.mjs
@@ -144,6 +144,7 @@ export default defineConfig({
             { label: "Pinecone", slug: "plugins/pinecone" },
             { label: "Chroma", slug: "plugins/chroma" },
             { label: "Qdrant", slug: "plugins/qdrant" },
+            { label: "Milvus", slug: "plugins/milvus" },
             { label: "MongoDB", slug: "plugins/mongodb" },
           ],
         },
diff --git a/docs/src/content/docs/index.mdx b/docs/src/content/docs/index.mdx
index 76d4f5336..6f4b1d721 100644
--- a/docs/src/content/docs/index.mdx
+++ b/docs/src/content/docs/index.mdx
@@ -270,6 +270,7 @@ String reply = chat.send("What is my name?").text(); // "Your name is Ada Lovela
   
   
   
+  
   
   
 
diff --git a/docs/src/content/docs/plugins/milvus.md b/docs/src/content/docs/plugins/milvus.md
new file mode 100644
index 000000000..09a97a892
--- /dev/null
+++ b/docs/src/content/docs/plugins/milvus.md
@@ -0,0 +1,72 @@
+---
+title: Milvus
+description: Milvus vector database integration for Genkit RAG workflows.
+---
+
+The Milvus plugin registers retrievers and indexers backed by a [Milvus](https://milvus.io/) server (v2 REST API) for Retrieval-Augmented Generation.
+
+## Installation
+
+```xml
+
+    com.google.genkit
+    genkit-plugin-milvus
+    1.0.0-SNAPSHOT
+
+```
+
+## Requirements
+
+- A running Milvus server. For local development:
+  ```bash
+  curl -sfL https://raw.githubusercontent.com/milvus-io/milvus/master/scripts/standalone_embed.sh -o standalone_embed.sh
+  bash standalone_embed.sh start
+  ```
+- Java 21+
+- An embedder (e.g. from the Google GenAI plugin)
+
+## Usage
+
+`MilvusPlugin` registers a retriever and indexer named `milvus/` for each configured collection.
+
+```java
+import com.google.genkit.plugins.milvus.MilvusCollectionConfig;
+import com.google.genkit.plugins.milvus.MilvusPlugin;
+
+Genkit genkit = Genkit.builder()
+    .plugin(GoogleGenAIPlugin.create(apiKey))
+    .plugin(
+        MilvusPlugin.builder()
+            .url("http://localhost:19530")   // default; serves both gRPC and REST
+            .token(System.getenv("MILVUS_TOKEN")) // optional; required for auth-enabled servers / Zilliz Cloud
+            .addCollection(
+                MilvusCollectionConfig.builder()
+                    .collectionName("films")
+                    .embedderName("googleai/gemini-embedding-001")
+                    .metric(MilvusCollectionConfig.Metric.COSINE)
+                    .createCollectionIfNotExists(true) // default
+                    .build())
+            .build())
+    .build();
+
+// Index and retrieve
+genkit.index("milvus/films", documents);
+List results = genkit.retrieve("milvus/films", "a Christopher Nolan sci-fi film");
+```
+
+## Configuration
+
+Tune per-collection settings with `MilvusCollectionConfig`:
+
+- `collectionName` — the Milvus collection name (required)
+- `embedderName` — the embedder used to vectorize documents and queries (required)
+- `dimension` — the embedding dimension used when creating the collection (default `768`). The store also probes the embedder on first use and creates the collection with the model's actual output dimension, so this is only a fallback.
+- `metric` — `COSINE` (default), `L2`, or `INNER_PRODUCT`
+- `createCollectionIfNotExists` — auto-create the collection (default `true`)
+- `addAdditionalMetadata(key, value)` — metadata merged into every indexed document
+
+Collections are created in Milvus "quick setup" mode (auto id primary key, a `vector` field, and dynamic fields). The document text is stored under `text` and its metadata as a JSON string under `metadata`.
+
+## Sample
+
+See the [milvus sample](https://github.com/genkit-ai/genkit-java/tree/main/samples/milvus).
diff --git a/plugins/milvus/pom.xml b/plugins/milvus/pom.xml
new file mode 100644
index 000000000..5ee3c1f6c
--- /dev/null
+++ b/plugins/milvus/pom.xml
@@ -0,0 +1,78 @@
+
+
+
+    4.0.0
+
+    
+        com.google.genkit
+        genkit-parent
+        1.0.0-SNAPSHOT
+        ../../pom.xml
+    
+
+    genkit-plugin-milvus
+    jar
+    Genkit Milvus Plugin
+    Milvus vector database integration for Genkit - indexer and retriever for RAG workflows
+
+    
+        false
+    
+
+    
+        
+        
+            com.google.genkit
+            genkit-core
+            ${project.version}
+        
+        
+            com.google.genkit
+            genkit-ai
+            ${project.version}
+        
+
+        
+        
+            com.fasterxml.jackson.core
+            jackson-databind
+        
+
+        
+        
+            org.slf4j
+            slf4j-api
+        
+
+        
+        
+            org.junit.jupiter
+            junit-jupiter
+            test
+        
+        
+            org.mockito
+            mockito-core
+            test
+        
+    
+
diff --git a/plugins/milvus/src/main/java/com/google/genkit/plugins/milvus/MilvusCollectionConfig.java b/plugins/milvus/src/main/java/com/google/genkit/plugins/milvus/MilvusCollectionConfig.java
new file mode 100644
index 000000000..2f6468c85
--- /dev/null
+++ b/plugins/milvus/src/main/java/com/google/genkit/plugins/milvus/MilvusCollectionConfig.java
@@ -0,0 +1,235 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package com.google.genkit.plugins.milvus;
+
+import java.util.HashMap;
+import java.util.Map;
+
+/**
+ * Configuration for a single Milvus collection managed by {@link MilvusPlugin}.
+ *
+ * 
Each config registers a retriever and indexer named {@code milvus/}. The
+ * collection is created in Milvus "quick setup" mode (auto id primary key, a {@code vector} field,
+ * and dynamic fields) storing the document text under {@code text} and its metadata as a JSON
+ * string under {@code metadata}.
+ */
+public final class MilvusCollectionConfig {
+
+  /** Vector similarity metric used by the Milvus index. */
+  public enum Metric {
+    COSINE("COSINE"),
+    L2("L2"),
+    INNER_PRODUCT("IP");
+
+    private final String value;
+
+    Metric(String value) {
+      this.value = value;
+    }
+
+    /**
+     * Returns the Milvus metric type name.
+     *
+     * @return the metric type name
+     */
+    public String getValue() {
+      return value;
+    }
+  }
+
+  private final String collectionName;
+  private final String embedderName;
+  private final int dimension;
+  private final Metric metric;
+  private final boolean createCollectionIfNotExists;
+  private final Map additionalMetadata;
+
+  private MilvusCollectionConfig(Builder builder) {
+    this.collectionName = builder.collectionName;
+    this.embedderName = builder.embedderName;
+    this.dimension = builder.dimension;
+    this.metric = builder.metric;
+    this.createCollectionIfNotExists = builder.createCollectionIfNotExists;
+    this.additionalMetadata = new HashMap<>(builder.additionalMetadata);
+  }
+
+  /**
+   * Returns the Milvus collection name.
+   *
+   * @return the collection name
+   */
+  public String getCollectionName() {
+    return collectionName;
+  }
+
+  /**
+   * Returns the name of the embedder used to vectorize documents and queries.
+   *
+   * @return the embedder name
+   */
+  public String getEmbedderName() {
+    return embedderName;
+  }
+
+  /**
+   * Returns the embedding dimension (default {@code 768}).
+   *
+   * @return the embedding dimension
+   */
+  public int getDimension() {
+    return dimension;
+  }
+
+  /**
+   * Returns the vector similarity metric (default {@link Metric#COSINE}).
+   *
+   * @return the metric
+   */
+  public Metric getMetric() {
+    return metric;
+  }
+
+  /**
+   * Returns whether to create the collection on first use if it does not exist (default {@code
+   * true}).
+   *
+   * @return {@code true} if the collection should be created when missing
+   */
+  public boolean isCreateCollectionIfNotExists() {
+    return createCollectionIfNotExists;
+  }
+
+  /**
+   * Returns additional metadata merged into every indexed document.
+   *
+   * @return the additional metadata
+   */
+  public Map getAdditionalMetadata() {
+    return additionalMetadata;
+  }
+
+  /**
+   * Creates a new builder.
+   *
+   * @return a new builder
+   */
+  public static Builder builder() {
+    return new Builder();
+  }
+
+  /** Builder for {@link MilvusCollectionConfig}. */
+  public static final class Builder {
+    private String collectionName;
+    private String embedderName;
+    private int dimension = 768;
+    private Metric metric = Metric.COSINE;
+    private boolean createCollectionIfNotExists = true;
+    private final Map additionalMetadata = new HashMap<>();
+
+    private Builder() {}
+
+    /**
+     * Sets the collection name.
+     *
+     * @param collectionName the collection name
+     * @return this builder
+     */
+    public Builder collectionName(String collectionName) {
+      this.collectionName = collectionName;
+      return this;
+    }
+
+    /**
+     * Sets the embedder name.
+     *
+     * @param embedderName the embedder name
+     * @return this builder
+     */
+    public Builder embedderName(String embedderName) {
+      this.embedderName = embedderName;
+      return this;
+    }
+
+    /**
+     * Sets the embedding dimension.
+     *
+     * @param dimension the embedding dimension (must be {@code >= 1})
+     * @return this builder
+     */
+    public Builder dimension(int dimension) {
+      this.dimension = dimension;
+      return this;
+    }
+
+    /**
+     * Sets the vector similarity metric.
+     *
+     * @param metric the metric
+     * @return this builder
+     */
+    public Builder metric(Metric metric) {
+      this.metric = metric;
+      return this;
+    }
+
+    /**
+     * Sets whether to create the collection on first use if it does not exist.
+     *
+     * @param createCollectionIfNotExists whether to create the collection when missing
+     * @return this builder
+     */
+    public Builder createCollectionIfNotExists(boolean createCollectionIfNotExists) {
+      this.createCollectionIfNotExists = createCollectionIfNotExists;
+      return this;
+    }
+
+    /**
+     * Adds a metadata entry merged into every indexed document.
+     *
+     * @param key the metadata key
+     * @param value the metadata value
+     * @return this builder
+     */
+    public Builder addAdditionalMetadata(String key, Object value) {
+      this.additionalMetadata.put(key, value);
+      return this;
+    }
+
+    /**
+     * Builds a new {@code MilvusCollectionConfig}.
+     *
+     * @return a new config instance
+     */
+    public MilvusCollectionConfig build() {
+      if (collectionName == null || collectionName.isBlank()) {
+        throw new IllegalArgumentException("collectionName must be non-empty");
+      }
+      if (embedderName == null || embedderName.isBlank()) {
+        throw new IllegalArgumentException("embedderName must be non-empty");
+      }
+      if (dimension < 1) {
+        throw new IllegalArgumentException("dimension must be >= 1");
+      }
+      if (metric == null) {
+        throw new IllegalArgumentException("metric must be non-null");
+      }
+      return new MilvusCollectionConfig(this);
+    }
+  }
+}
diff --git a/plugins/milvus/src/main/java/com/google/genkit/plugins/milvus/MilvusPlugin.java b/plugins/milvus/src/main/java/com/google/genkit/plugins/milvus/MilvusPlugin.java
new file mode 100644
index 000000000..f14aa3bb8
--- /dev/null
+++ b/plugins/milvus/src/main/java/com/google/genkit/plugins/milvus/MilvusPlugin.java
@@ -0,0 +1,175 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package com.google.genkit.plugins.milvus;
+
+import com.google.genkit.ai.Embedder;
+import com.google.genkit.core.Action;
+import com.google.genkit.core.ActionType;
+import com.google.genkit.core.Plugin;
+import com.google.genkit.core.Registry;
+import java.util.ArrayList;
+import java.util.List;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+/**
+ * Milvus vector database plugin for Genkit.
+ *
+ * 
Registers a retriever and indexer named {@code milvus/} for each configured
+ * collection, talking to a Milvus server over its v2 REST API.
+ *
+ * 
Example usage:
+ *
+ * 
{@code
+ * Genkit genkit = Genkit.builder()
+ *     .plugin(GoogleGenAIPlugin.create(apiKey))
+ *     .plugin(
+ *         MilvusPlugin.builder()
+ *             .url("http://localhost:19530")
+ *             .addCollection(
+ *                 MilvusCollectionConfig.builder()
+ *                     .collectionName("films")
+ *                     .embedderName("googleai/gemini-embedding-001")
+ *                     .build())
+ *             .build())
+ *     .build();
+ * }

+ */
+public final class MilvusPlugin implements Plugin {
+
+  /** The plugin name; used as the {@code milvus/...} action prefix. */
+  public static final String PLUGIN_NAME = "milvus";
+
+  /** Default Milvus server URL. */
+  public static final String DEFAULT_URL = "http://localhost:19530";
+
+  private static final Logger logger = LoggerFactory.getLogger(MilvusPlugin.class);
+
+  private final String url;
+  private final String token;
+  private final List collectionConfigs;
+
+  private MilvusPlugin(Builder builder) {
+    this.url = builder.url;
+    this.token = builder.token;
+    this.collectionConfigs = new ArrayList<>(builder.collectionConfigs);
+  }
+
+  /**
+   * Creates a new builder.
+   *
+   * @return a new builder
+   */
+  public static Builder builder() {
+    return new Builder();
+  }
+
+  @Override
+  public String getName() {
+    return PLUGIN_NAME;
+  }
+
+  @Override
+  public List> init() {
+    throw new IllegalStateException(
+        "MilvusPlugin requires a Registry to resolve embedders. Use init(registry) instead.");
+  }
+
+  @Override
+  public List> init(Registry registry) {
+    List> actions = new ArrayList<>();
+    for (MilvusCollectionConfig config : collectionConfigs) {
+      String embedderKey = ActionType.EMBEDDER.keyFromName(config.getEmbedderName());
+      Action embedderAction = registry.lookupAction(embedderKey);
+      if (embedderAction == null) {
+        throw new IllegalStateException(
+            "Embedder not found: "
+                + config.getEmbedderName()
+                + ". Make sure the embedder plugin is registered before MilvusPlugin.");
+      }
+      if (!(embedderAction instanceof Embedder embedder)) {
+        throw new IllegalStateException(
+            "Action " + config.getEmbedderName() + " is not an Embedder");
+      }
+
+      MilvusVectorStore store = new MilvusVectorStore(url, token, config, embedder);
+      actions.add(store.createRetriever());
+      actions.add(store.createIndexer());
+      logger.info("Registered Milvus vector store: {}/{}", PLUGIN_NAME, config.getCollectionName());
+    }
+    return actions;
+  }
+
+  /** Builder for {@link MilvusPlugin}. */
+  public static final class Builder {
+    private String url = DEFAULT_URL;
+    private String token;
+    private final List collectionConfigs = new ArrayList<>();
+
+    private Builder() {}
+
+    /**
+     * Sets the Milvus server URL (default {@value #DEFAULT_URL}).
+     *
+     * @param url the server URL
+     * @return this builder
+     */
+    public Builder url(String url) {
+      this.url = url;
+      return this;
+    }
+
+    /**
+     * Sets the Milvus auth token (optional; required for Zilliz Cloud or auth-enabled servers).
+     *
+     * @param token the auth token
+     * @return this builder
+     */
+    public Builder token(String token) {
+      this.token = token;
+      return this;
+    }
+
+    /**
+     * Adds a collection configuration.
+     *
+     * @param config the collection configuration
+     * @return this builder
+     */
+    public Builder addCollection(MilvusCollectionConfig config) {
+      this.collectionConfigs.add(config);
+      return this;
+    }
+
+    /**
+     * Builds the plugin.
+     *
+     * @return a new {@code MilvusPlugin}
+     */
+    public MilvusPlugin build() {
+      if (url == null || url.isBlank()) {
+        throw new IllegalStateException("url is required");
+      }
+      if (collectionConfigs.isEmpty()) {
+        throw new IllegalStateException("At least one collection configuration is required");
+      }
+      return new MilvusPlugin(this);
+    }
+  }
+}
diff --git a/plugins/milvus/src/main/java/com/google/genkit/plugins/milvus/MilvusVectorStore.java b/plugins/milvus/src/main/java/com/google/genkit/plugins/milvus/MilvusVectorStore.java
new file mode 100644
index 000000000..222cba462
--- /dev/null
+++ b/plugins/milvus/src/main/java/com/google/genkit/plugins/milvus/MilvusVectorStore.java
@@ -0,0 +1,321 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package com.google.genkit.plugins.milvus;
+
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.fasterxml.jackson.databind.node.ArrayNode;
+import com.fasterxml.jackson.databind.node.ObjectNode;
+import com.google.genkit.ai.Document;
+import com.google.genkit.ai.EmbedRequest;
+import com.google.genkit.ai.EmbedResponse;
+import com.google.genkit.ai.Embedder;
+import com.google.genkit.ai.Indexer;
+import com.google.genkit.ai.IndexerRequest;
+import com.google.genkit.ai.IndexerResponse;
+import com.google.genkit.ai.Retriever;
+import com.google.genkit.ai.RetrieverRequest;
+import com.google.genkit.ai.RetrieverResponse;
+import com.google.genkit.core.ActionContext;
+import java.net.URI;
+import java.net.http.HttpClient;
+import java.net.http.HttpRequest;
+import java.net.http.HttpResponse;
+import java.nio.charset.StandardCharsets;
+import java.util.ArrayList;
+import java.util.HashMap;
+import java.util.List;
+import java.util.Map;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+/**
+ * Milvus vector store backed by the Milvus v2 REST API.
+ *
+ * 
Indexes documents into a Milvus collection (quick-setup mode: auto id, a {@code vector} field,
+ * dynamic {@code text} and {@code metadata} fields) and retrieves the nearest neighbors of a query
+ * embedding.
+ */
+public final class MilvusVectorStore {
+
+  private static final Logger logger = LoggerFactory.getLogger(MilvusVectorStore.class);
+  private static final ObjectMapper MAPPER = new ObjectMapper();
+  private static final String TEXT_FIELD = "text";
+  private static final String METADATA_FIELD = "metadata";
+  private static final String VECTOR_FIELD = "vector";
+
+  private final HttpClient http = HttpClient.newHttpClient();
+  private final String baseUrl;
+  private final String token;
+  private final MilvusCollectionConfig config;
+  private final Embedder embedder;
+
+  private volatile boolean initialized = false;
+
+  /**
+   * Creates a new store.
+   *
+   * @param baseUrl the Milvus server base URL (e.g. {@code http://localhost:19530})
+   * @param token the Milvus auth token, or {@code null} when the server requires none
+   * @param config the collection configuration
+   * @param embedder the embedder used to vectorize documents and queries
+   */
+  public MilvusVectorStore(
+      String baseUrl, String token, MilvusCollectionConfig config, Embedder embedder) {
+    this.baseUrl = baseUrl.endsWith("/") ? baseUrl.substring(0, baseUrl.length() - 1) : baseUrl;
+    this.token = token;
+    this.config = config;
+    this.embedder = embedder;
+  }
+
+  /** Creates the retriever action registered by the plugin. */
+  Retriever createRetriever() {
+    String name = MilvusPlugin.PLUGIN_NAME + "/" + config.getCollectionName();
+    return Retriever.builder().name(name).handler(this::retrieve).build();
+  }
+
+  /** Creates the indexer action registered by the plugin. */
+  Indexer createIndexer() {
+    String name = MilvusPlugin.PLUGIN_NAME + "/" + config.getCollectionName();
+    return Indexer.builder().name(name).handler(this::index).build();
+  }
+
+  private synchronized void ensureInitialized() {
+    if (initialized) {
+      return;
+    }
+    if (config.isCreateCollectionIfNotExists() && !collectionExists()) {
+      ObjectNode body = MAPPER.createObjectNode();
+      body.put("collectionName", config.getCollectionName());
+      body.put("dimension", resolveDimension());
+      body.put("metricType", config.getMetric().getValue());
+      // Let Milvus generate the Int64 primary key so callers don't have to supply one.
+      body.put("autoID", true);
+      send("/v2/vectordb/collections/create", body);
+      logger.info("Created Milvus collection {}", config.getCollectionName());
+    }
+    initialized = true;
+  }
+
+  private boolean collectionExists() {
+    ObjectNode body = MAPPER.createObjectNode();
+    body.put("collectionName", config.getCollectionName());
+    JsonNode data = send("/v2/vectordb/collections/has", body).get("data");
+    return data != null && data.has("has") && data.get("has").asBoolean();
+  }
+
+  private int resolveDimension() {
+    try {
+      return generateEmbedding(null, "genkit dimension probe").size();
+    } catch (RuntimeException e) {
+      logger.debug(
+          "Embedding probe failed; using configured dimension {}: {}",
+          config.getDimension(),
+          e.getMessage());
+      return config.getDimension();
+    }
+  }
+
+  /**
+   * Retrieves documents similar to the query.
+   *
+   * @param context the action context
+   * @param request the retriever request
+   * @return the retriever response with matching documents
+   */
+  public RetrieverResponse retrieve(ActionContext context, RetrieverRequest request) {
+    ensureInitialized();
+    Document queryDoc = request.getQuery();
+    if (queryDoc == null || queryDoc.text() == null || queryDoc.text().isBlank()) {
+      throw new RuntimeException("Query document has no text content");
+    }
+    int topK =
+        request.getOptions() != null && request.getOptions().getK() != null
+            ? request.getOptions().getK()
+            : 10;
+    List queryEmbedding = generateEmbedding(context, queryDoc.text());
+
+    ObjectNode body = MAPPER.createObjectNode();
+    body.put("collectionName", config.getCollectionName());
+    body.put("annsField", VECTOR_FIELD);
+    body.put("limit", topK);
+    ArrayNode data = body.putArray("data");
+    data.add(floatsToArray(queryEmbedding));
+    ArrayNode outputFields = body.putArray("outputFields");
+    outputFields.add(TEXT_FIELD);
+    outputFields.add(METADATA_FIELD);
+
+    JsonNode resp = send("/v2/vectordb/entities/search", body);
+    List documents = new ArrayList<>();
+    JsonNode results = resp.get("data");
+    if (results != null && results.isArray()) {
+      for (JsonNode hit : results) {
+        documents.add(toDocument(hit));
+      }
+    }
+    logger.debug(
+        "Retrieved {} documents from collection {}", documents.size(), config.getCollectionName());
+    return new RetrieverResponse(documents);
+  }
+
+  /**
+   * Indexes documents into the collection, generating an embedding for each.
+   *
+   * @param context the action context
+   * @param request the indexer request
+   * @return the indexer response
+   */
+  public IndexerResponse index(ActionContext context, IndexerRequest request) {
+    ensureInitialized();
+    List documents = request.getDocuments();
+    if (documents == null || documents.isEmpty()) {
+      logger.warn("No documents to index");
+      return new IndexerResponse();
+    }
+
+    ObjectNode body = MAPPER.createObjectNode();
+    body.put("collectionName", config.getCollectionName());
+    ArrayNode data = body.putArray("data");
+    for (Document doc : documents) {
+      String content = doc.text() != null ? doc.text() : "";
+      List embedding = generateEmbedding(context, content);
+
+      ObjectNode entity = data.addObject();
+      entity.set(VECTOR_FIELD, floatsToArray(embedding));
+      entity.put(TEXT_FIELD, content);
+
+      Map metadata = new HashMap<>();
+      if (doc.getMetadata() != null) {
+        for (Map.Entry entry : doc.getMetadata().entrySet()) {
+          if (!"id".equals(entry.getKey())) {
+            metadata.put(entry.getKey(), entry.getValue());
+          }
+        }
+      }
+      metadata.putAll(config.getAdditionalMetadata());
+      entity.put(METADATA_FIELD, writeJson(metadata));
+    }
+
+    send("/v2/vectordb/entities/insert", body);
+    logger.info(
+        "Indexed {} documents into collection {}", documents.size(), config.getCollectionName());
+    return new IndexerResponse();
+  }
+
+  private Document toDocument(JsonNode hit) {
+    Map metadata = new HashMap<>();
+    String content = "";
+    JsonNode textNode = hit.get(TEXT_FIELD);
+    if (textNode != null && !textNode.isNull()) {
+      content = textNode.asText();
+    }
+    JsonNode metadataNode = hit.get(METADATA_FIELD);
+    if (metadataNode != null && !metadataNode.isNull()) {
+      try {
+        String raw = metadataNode.isTextual() ? metadataNode.asText() : metadataNode.toString();
+        if (raw != null && !raw.isBlank()) {
+          @SuppressWarnings("unchecked")
+          Map parsed = MAPPER.readValue(raw, Map.class);
+          metadata.putAll(parsed);
+        }
+      } catch (Exception e) {
+        logger.debug("Failed to parse Milvus metadata: {}", e.getMessage());
+      }
+    }
+    if (hit.get("id") != null) {
+      metadata.put("id", hit.get("id").asText());
+    }
+    if (hit.get("distance") != null && hit.get("distance").isNumber()) {
+      double distance = hit.get("distance").asDouble();
+      metadata.put("distance", distance);
+      metadata.put("score", distance);
+    }
+    Document doc = new Document(content);
+    doc.setMetadata(metadata);
+    return doc;
+  }
+
+  private ArrayNode floatsToArray(List values) {
+    ArrayNode arr = MAPPER.createArrayNode();
+    for (float v : values) {
+      arr.add(v);
+    }
+    return arr;
+  }
+
+  private List generateEmbedding(ActionContext ctx, String text) {
+    EmbedResponse response = embedder.run(ctx, new EmbedRequest(List.of(new Document(text))));
+    if (response.getEmbeddings() == null || response.getEmbeddings().isEmpty()) {
+      throw new RuntimeException("Failed to generate embedding for text");
+    }
+    float[] values = response.getEmbeddings().get(0).getValues();
+    List out = new ArrayList<>(values.length);
+    for (float v : values) {
+      out.add(v);
+    }
+    return out;
+  }
+
+  private static String writeJson(Object value) {
+    try {
+      return MAPPER.writeValueAsString(value);
+    } catch (Exception e) {
+      throw new RuntimeException("Failed to serialize metadata: " + e.getMessage(), e);
+    }
+  }
+
+  /** Posts a JSON request to the Milvus REST API and validates both the HTTP and Milvus codes. */
+  private JsonNode send(String path, JsonNode body) {
+    try {
+      HttpRequest.Builder builder =
+          HttpRequest.newBuilder()
+              .uri(URI.create(baseUrl + path))
+              .header("Content-Type", "application/json")
+              .header("Accept", "application/json");
+      if (token != null && !token.isBlank()) {
+        builder.header("Authorization", "Bearer " + token);
+      }
+      builder.POST(
+          HttpRequest.BodyPublishers.ofString(
+              MAPPER.writeValueAsString(body), StandardCharsets.UTF_8));
+      HttpResponse response =
+          http.send(builder.build(), HttpResponse.BodyHandlers.ofString());
+      if (response.statusCode() / 100 != 2) {
+        throw new RuntimeException(
+            "Milvus request "
+                + path
+                + " failed ("
+                + response.statusCode()
+                + "): "
+                + response.body());
+      }
+      JsonNode node = MAPPER.readTree(response.body());
+      JsonNode code = node.get("code");
+      if (code != null && code.asInt() != 0) {
+        throw new RuntimeException(
+            "Milvus request " + path + " failed (code " + code.asInt() + "): " + node.toString());
+      }
+      return node;
+    } catch (RuntimeException e) {
+      throw e;
+    } catch (Exception e) {
+      throw new RuntimeException("Milvus request " + path + " failed: " + e.getMessage(), e);
+    }
+  }
+}
diff --git a/plugins/milvus/src/main/java/com/google/genkit/plugins/milvus/package-info.java b/plugins/milvus/src/main/java/com/google/genkit/plugins/milvus/package-info.java
new file mode 100644
index 000000000..7f39cda73
--- /dev/null
+++ b/plugins/milvus/src/main/java/com/google/genkit/plugins/milvus/package-info.java
@@ -0,0 +1,25 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+/**
+ * Milvus vector database integration for Genkit.
+ *
+ * 
{@link com.google.genkit.plugins.milvus.MilvusPlugin} registers retrievers and indexers backed
+ * by a Milvus server (v2 REST API) for RAG workflows.
+ */
+package com.google.genkit.plugins.milvus;
diff --git a/plugins/milvus/src/test/java/com/google/genkit/plugins/milvus/MilvusCollectionConfigTest.java b/plugins/milvus/src/test/java/com/google/genkit/plugins/milvus/MilvusCollectionConfigTest.java
new file mode 100644
index 000000000..b78cfe69d
--- /dev/null
+++ b/plugins/milvus/src/test/java/com/google/genkit/plugins/milvus/MilvusCollectionConfigTest.java
@@ -0,0 +1,75 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package com.google.genkit.plugins.milvus;
+
+import static org.junit.jupiter.api.Assertions.assertEquals;
+import static org.junit.jupiter.api.Assertions.assertThrows;
+import static org.junit.jupiter.api.Assertions.assertTrue;
+
+import org.junit.jupiter.api.Test;
+
+/** Unit tests for {@link MilvusCollectionConfig}. */
+class MilvusCollectionConfigTest {
+
+  @Test
+  void defaultsAreSane() {
+    MilvusCollectionConfig c =
+        MilvusCollectionConfig.builder().collectionName("films").embedderName("e").build();
+    assertEquals("films", c.getCollectionName());
+    assertEquals(768, c.getDimension());
+    assertEquals(MilvusCollectionConfig.Metric.COSINE, c.getMetric());
+    assertEquals("COSINE", c.getMetric().getValue());
+    assertTrue(c.isCreateCollectionIfNotExists());
+  }
+
+  @Test
+  void customBuilder() {
+    MilvusCollectionConfig c =
+        MilvusCollectionConfig.builder()
+            .collectionName("docs")
+            .embedderName("e")
+            .dimension(1536)
+            .metric(MilvusCollectionConfig.Metric.INNER_PRODUCT)
+            .createCollectionIfNotExists(false)
+            .addAdditionalMetadata("source", "wiki")
+            .build();
+    assertEquals(1536, c.getDimension());
+    assertEquals("IP", c.getMetric().getValue());
+    assertEquals(false, c.isCreateCollectionIfNotExists());
+    assertEquals("wiki", c.getAdditionalMetadata().get("source"));
+  }
+
+  @Test
+  void builderValidates() {
+    assertThrows(
+        IllegalArgumentException.class,
+        () -> MilvusCollectionConfig.builder().embedderName("e").build());
+    assertThrows(
+        IllegalArgumentException.class,
+        () -> MilvusCollectionConfig.builder().collectionName("c").build());
+    assertThrows(
+        IllegalArgumentException.class,
+        () ->
+            MilvusCollectionConfig.builder()
+                .collectionName("c")
+                .embedderName("e")
+                .dimension(0)
+                .build());
+  }
+}
diff --git a/plugins/milvus/src/test/java/com/google/genkit/plugins/milvus/MilvusPluginTest.java b/plugins/milvus/src/test/java/com/google/genkit/plugins/milvus/MilvusPluginTest.java
new file mode 100644
index 000000000..a165d488d
--- /dev/null
+++ b/plugins/milvus/src/test/java/com/google/genkit/plugins/milvus/MilvusPluginTest.java
@@ -0,0 +1,48 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package com.google.genkit.plugins.milvus;
+
+import static org.junit.jupiter.api.Assertions.assertEquals;
+import static org.junit.jupiter.api.Assertions.assertThrows;
+
+import org.junit.jupiter.api.Test;
+
+/** Unit tests for {@link MilvusPlugin}. */
+class MilvusPluginTest {
+
+  private static MilvusCollectionConfig config() {
+    return MilvusCollectionConfig.builder().collectionName("films").embedderName("e").build();
+  }
+
+  @Test
+  void getName() {
+    assertEquals("milvus", MilvusPlugin.builder().addCollection(config()).build().getName());
+  }
+
+  @Test
+  void requiresAtLeastOneCollection() {
+    assertThrows(IllegalStateException.class, () -> MilvusPlugin.builder().build());
+  }
+
+  @Test
+  void initWithoutRegistryThrows() {
+    MilvusPlugin plugin = MilvusPlugin.builder().addCollection(config()).build();
+    assertThrows(IllegalStateException.class, plugin::init);
+  }
+}
diff --git a/plugins/milvus/src/test/java/com/google/genkit/plugins/milvus/MilvusVectorStoreTest.java b/plugins/milvus/src/test/java/com/google/genkit/plugins/milvus/MilvusVectorStoreTest.java
new file mode 100644
index 000000000..317383194
--- /dev/null
+++ b/plugins/milvus/src/test/java/com/google/genkit/plugins/milvus/MilvusVectorStoreTest.java
@@ -0,0 +1,118 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package com.google.genkit.plugins.milvus;
+
+import static org.junit.jupiter.api.Assertions.assertEquals;
+import static org.junit.jupiter.api.Assertions.assertFalse;
+import static org.junit.jupiter.api.Assumptions.assumeTrue;
+
+import com.google.genkit.ai.Document;
+import com.google.genkit.ai.EmbedResponse;
+import com.google.genkit.ai.Embedder;
+import com.google.genkit.ai.EmbedderInfo;
+import com.google.genkit.ai.IndexerRequest;
+import com.google.genkit.ai.RetrieverRequest;
+import com.google.genkit.ai.RetrieverResponse;
+import java.util.ArrayList;
+import java.util.List;
+import java.util.Random;
+import java.util.UUID;
+import org.junit.jupiter.api.BeforeEach;
+import org.junit.jupiter.api.Test;
+
+/**
+ * Integration tests for {@link MilvusVectorStore}, gated on the {@code MILVUS_URL} environment
+ * variable (e.g. a Milvus server started with the standalone Docker script). Skipped via {@link
+ * org.junit.jupiter.api.Assumptions} when unset. Uses a deterministic stub embedder so a query
+ * equal to an indexed document's text retrieves that document first.
+ */
+class MilvusVectorStoreTest {
+
+  private static final String URL = System.getenv("MILVUS_URL");
+  private static final String TOKEN = System.getenv("MILVUS_TOKEN");
+  private static final int DIM = 16;
+
+  private MilvusVectorStore store;
+
+  private static boolean configured() {
+    return URL != null && !URL.isEmpty();
+  }
+
+  private static Embedder stubEmbedder() {
+    return new Embedder(
+        "test/stub",
+        new EmbedderInfo(),
+        (ctx, req) -> {
+          List out = new ArrayList<>();
+          for (Document doc : req.getDocuments()) {
+            Random random = new Random(doc.text().hashCode());
+            float[] values = new float[DIM];
+            for (int i = 0; i < DIM; i++) {
+              values[i] = random.nextFloat();
+            }
+            out.add(new EmbedResponse.Embedding(values));
+          }
+          return new EmbedResponse(out);
+        });
+  }
+
+  @BeforeEach
+  void setUp() {
+    if (!configured()) {
+      return;
+    }
+    MilvusCollectionConfig config =
+        MilvusCollectionConfig.builder()
+            .collectionName("genkit_test_" + UUID.randomUUID().toString().replace("-", ""))
+            .embedderName("test/stub")
+            .dimension(DIM)
+            .build();
+    store = new MilvusVectorStore(URL, TOKEN, config, stubEmbedder());
+  }
+
+  @Test
+  void indexThenRetrieveReturnsNearestFirst() throws Exception {
+    assumeTrue(configured());
+    List docs =
+        List.of(
+            Document.fromText("The Matrix is a sci-fi film about simulated reality."),
+            Document.fromText("The Godfather is a crime film about a mafia family."),
+            Document.fromText("Inception is a sci-fi film about dreams."));
+    store.index(null, new IndexerRequest(docs));
+
+    RetrieverRequest request = new RetrieverRequest(Document.fromText(docs.get(1).text()));
+    RetrieverRequest.RetrieverOptions options = new RetrieverRequest.RetrieverOptions();
+    options.setK(3);
+    request.setOptions(options);
+
+    // Newly inserted data may take a moment to become searchable.
+    RetrieverResponse response = null;
+    for (int attempt = 0; attempt < 15; attempt++) {
+      response = store.retrieve(null, request);
+      if (!response.getDocuments().isEmpty()) {
+        break;
+      }
+      Thread.sleep(1000);
+    }
+    assertFalse(response.getDocuments().isEmpty());
+    assertEquals(
+        "The Godfather is a crime film about a mafia family.",
+        response.getDocuments().get(0).text());
+  }
+}
diff --git a/pom.xml b/pom.xml
index e9a79e558..69fa99f2d 100644
--- a/pom.xml
+++ b/pom.xml
@@ -90,6 +90,7 @@
         plugins/mongodb
         plugins/chroma
         plugins/qdrant
+        plugins/milvus
         plugins/pinecone
         plugins/evaluators
         plugins/aws-bedrock
@@ -128,6 +129,7 @@
         samples/pinecone
         samples/chroma
         samples/qdrant
+        samples/milvus
         samples/mongo-vector
         samples/structured-output
         samples/aws-bedrock
diff --git a/samples/README.md b/samples/README.md
index 0e910b6ef..b14386858 100644
--- a/samples/README.md
+++ b/samples/README.md
@@ -79,6 +79,7 @@ The Dev UI will be available at `http://localhost:4000` and allows you to:
 | [pinecone](./pinecone) | Pinecone vector database RAG sample | `OPENAI_API_KEY` + `PINECONE_API_KEY` |
 | [chroma](./chroma) | Chroma vector database RAG sample | `GEMINI_API_KEY` + Chroma |
 | [qdrant](./qdrant) | Qdrant vector database RAG sample | `GEMINI_API_KEY` + Qdrant |
+| [milvus](./milvus) | Milvus vector database RAG sample | `GEMINI_API_KEY` + Milvus |
 | [mongo-vector](./mongo-vector) | MongoDB Atlas Vector Search RAG sample | `GEMINI_API_KEY` + MongoDB Atlas |
 | [agents-human-in-the-loop](./agents-human-in-the-loop) | Agent interrupts + human-in-the-loop resume | `GEMINI_API_KEY` |
 | [agents-firestore-session](./agents-firestore-session) | Agent session persistence backed by Firestore | `GEMINI_API_KEY` + Firestore |
diff --git a/samples/milvus/README.md b/samples/milvus/README.md
new file mode 100644
index 000000000..87384c7d2
--- /dev/null
+++ b/samples/milvus/README.md
@@ -0,0 +1,61 @@
+# Milvus RAG Sample
+
+A Retrieval-Augmented Generation sample that indexes film descriptions into **Milvus** via `MilvusPlugin` and answers questions with a **Gemini** model.
+
+## Prerequisites
+
+- Java 21+ and Maven 3.6+
+- A `GEMINI_API_KEY` — get one from [Google AI Studio](https://aistudio.google.com/apikey)
+- A reachable Milvus server (see the Docker command below)
+
+## Run Milvus locally with Docker
+
+Milvus ships a helper script that runs a single standalone container (with embedded etcd + local storage):
+
+```bash
+curl -sfL https://raw.githubusercontent.com/milvus-io/milvus/master/scripts/standalone_embed.sh -o standalone_embed.sh
+bash standalone_embed.sh start
+
+# wait until it's accepting connections (health probe on 9091)
+until curl -sf http://localhost:9091/healthz >/dev/null 2>&1; do sleep 1; done
+echo "milvus ready"
+```
+
+Milvus serves both gRPC and the REST API on port `19530`. The collection is created automatically on first use. Stop and remove it later with:
+
+```bash
+bash standalone_embed.sh stop && bash standalone_embed.sh delete
+```
+
+## Configure
+
+```bash
+export GEMINI_API_KEY=
+export MILVUS_URL=http://localhost:19530     # optional (default)
+export MILVUS_TOKEN=          # optional (required for auth-enabled servers / Zilliz Cloud)
+export MILVUS_COLLECTION=genkit_films        # optional (default)
+```
+
+## Run
+
+```bash
+mvn -q exec:java
+```
+
+Then open the Genkit Dev UI at http://localhost:4000 (or run under `genkit start -- mvn -q exec:java`) and exercise the flows:
+
+- `indexDocuments` — index the sample film descriptions
+- `retrieveDocuments` — return the films matching a query
+- `ragQuery` — answer a question using retrieved context
+
+Or via curl:
+
+```bash
+curl -X POST http://localhost:4000/api/flows/indexDocuments -H 'Content-Type: application/json' -d '{}'
+curl -X POST http://localhost:4000/api/flows/ragQuery -H 'Content-Type: application/json' \
+  -d '{"data": "What Christopher Nolan films are mentioned?"}'
+```
+
+## Configuration
+
+`MilvusPlugin` talks to the Milvus v2 REST API. Tune per-collection settings with `MilvusCollectionConfig` (collection name, embedder, dimension, metric — `COSINE`/`L2`/`INNER_PRODUCT`, `createCollectionIfNotExists`, additional metadata). Collections are created in quick-setup mode; the document text is stored under `text` and metadata as a JSON string under `metadata`.
diff --git a/samples/milvus/pom.xml b/samples/milvus/pom.xml
new file mode 100644
index 000000000..d840da867
--- /dev/null
+++ b/samples/milvus/pom.xml
@@ -0,0 +1,86 @@
+
+
+
+    4.0.0
+
+    
+        com.google.genkit
+        genkit-parent
+        1.0.0-SNAPSHOT
+        ../../pom.xml
+    
+
+    genkit-sample-milvus
+    jar
+    Genkit Milvus RAG Sample
+    Sample application demonstrating Milvus vector search with Genkit
+
+    
+        true
+        com.google.genkit.samples.milvus.MilvusRAGSample
+    
+
+    
+        
+            com.google.genkit
+            genkit
+            ${project.version}
+        
+        
+            com.google.genkit
+            genkit-plugin-milvus
+            ${project.version}
+        
+        
+            com.google.genkit
+            genkit-plugin-google-genai
+            ${project.version}
+        
+        
+            com.google.genkit
+            genkit-plugin-jetty
+            ${project.version}
+        
+        
+            ch.qos.logback
+            logback-classic
+        
+        
+            io.github.cdimascio
+            dotenv-java
+            3.2.0
+        
+    
+
+    
+        
+            
+                org.codehaus.mojo
+                exec-maven-plugin
+                3.6.3
+                
+                    ${exec.mainClass}
+                
+            
+        
+    
+
diff --git a/samples/milvus/run.sh b/samples/milvus/run.sh
new file mode 100755
index 000000000..d10fb3e2a
--- /dev/null
+++ b/samples/milvus/run.sh
@@ -0,0 +1,7 @@
+#!/bin/bash
+
+# Genkit Sample Runner
+# This script runs the sample application
+
+echo "Building and running the sample..."
+mvn clean compile exec:java
diff --git a/samples/milvus/src/main/java/com/google/genkit/samples/milvus/MilvusRAGSample.java b/samples/milvus/src/main/java/com/google/genkit/samples/milvus/MilvusRAGSample.java
new file mode 100644
index 000000000..4843a19e9
--- /dev/null
+++ b/samples/milvus/src/main/java/com/google/genkit/samples/milvus/MilvusRAGSample.java
@@ -0,0 +1,165 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package com.google.genkit.samples.milvus;
+
+import com.google.genkit.Genkit;
+import com.google.genkit.GenkitOptions;
+import com.google.genkit.ai.*;
+import com.google.genkit.core.Flow;
+import com.google.genkit.plugins.googlegenai.GoogleGenAIPlugin;
+import com.google.genkit.plugins.jetty.JettyPlugin;
+import com.google.genkit.plugins.jetty.JettyPluginOptions;
+import com.google.genkit.plugins.milvus.MilvusCollectionConfig;
+import com.google.genkit.plugins.milvus.MilvusPlugin;
+import io.github.cdimascio.dotenv.Dotenv;
+import java.util.List;
+import java.util.stream.Collectors;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+/**
+ * Sample application demonstrating Milvus vector search with Genkit.
+ *
+ * 
Indexes a handful of film descriptions into a Milvus collection, retrieves the nearest matches
+ * for a query, and answers questions with a RAG flow. Requires a running Milvus server (see the
+ * README for a one-line Docker command) and a {@code GEMINI_API_KEY}.
+ */
+public class MilvusRAGSample {
+
+  private static final Logger logger = LoggerFactory.getLogger(MilvusRAGSample.class);
+
+  private static final List SAMPLE_DOCUMENTS =
+      List.of(
+          "The Godfather is a 1972 crime film directed by Francis Ford Coppola about the Corleone crime family.",
+          "The Dark Knight is a 2008 superhero film directed by Christopher Nolan featuring Batman against the Joker.",
+          "Pulp Fiction is a 1994 crime film directed by Quentin Tarantino known for its nonlinear narrative.",
+          "Inception is a 2010 sci-fi film directed by Christopher Nolan about dream infiltration.",
+          "The Matrix is a 1999 sci-fi film directed by the Wachowskis exploring simulated reality.",
+          "Forrest Gump is a 1994 drama directed by Robert Zemeckis about a man's extraordinary life.",
+          "Star Wars is a 1977 sci-fi film directed by George Lucas set in a galaxy far, far away.",
+          "The Shawshank Redemption is a 1994 drama about hope and friendship in a prison.");
+
+  private static final String RAG_SYSTEM_PROMPT =
+      """
+      You are a helpful assistant that answers questions based on the provided context documents.
+      Answer only from the context. If the context is insufficient, say so.
+      """;
+
+  public static void main(String[] args) {
+    Dotenv dotenv = Dotenv.configure().ignoreIfMissing().systemProperties().load();
+
+    String geminiApiKey = getEnv(dotenv, "GEMINI_API_KEY");
+    if (geminiApiKey == null) {
+      logger.error("Please set GEMINI_API_KEY in .env file or environment variable");
+      System.exit(1);
+    }
+
+    String milvusUrl = getEnvOrDefault(dotenv, "MILVUS_URL", "http://localhost:19530");
+    String milvusToken = getEnv(dotenv, "MILVUS_TOKEN"); // optional
+    String collection = getEnvOrDefault(dotenv, "MILVUS_COLLECTION", "genkit_films");
+
+    logger.info("Starting Milvus RAG Sample (url={}, collection={})", milvusUrl, collection);
+
+    MilvusPlugin milvusPlugin =
+        MilvusPlugin.builder()
+            .url(milvusUrl)
+            .token(milvusToken)
+            .addCollection(
+                MilvusCollectionConfig.builder()
+                    .collectionName(collection)
+                    .embedderName("googleai/gemini-embedding-001")
+                    .metric(MilvusCollectionConfig.Metric.COSINE)
+                    .build())
+            .build();
+
+    JettyPlugin jetty = new JettyPlugin(JettyPluginOptions.builder().port(8088).build());
+    Genkit genkit =
+        Genkit.builder()
+            .options(GenkitOptions.builder().devMode(true).reflectionPort(3100).build())
+            .plugin(GoogleGenAIPlugin.create(geminiApiKey))
+            .plugin(milvusPlugin)
+            .plugin(jetty)
+            .build();
+
+    String action = "milvus/" + collection;
+
+    Flow indexDocumentsFlow =
+        genkit.defineFlow(
+            "indexDocuments",
+            Void.class,
+            String.class,
+            (ctx, input) -> {
+              List documents =
+                  SAMPLE_DOCUMENTS.stream().map(Document::fromText).collect(Collectors.toList());
+              genkit.index(action, documents);
+              return "Successfully indexed " + documents.size() + " documents";
+            });
+
+    @SuppressWarnings("unchecked")
+    Flow, Void> retrieveDocumentsFlow =
+        genkit.defineFlow(
+            "retrieveDocuments",
+            String.class,
+            (Class>) (Class) List.class,
+            (ctx, query) -> {
+              List docs = genkit.retrieve(action, query);
+              return docs.stream().map(Document::text).collect(Collectors.toList());
+            });
+
+    Flow ragQueryFlow =
+        genkit.defineFlow(
+            "ragQuery",
+            String.class,
+            String.class,
+            (ctx, question) -> {
+              List docs = genkit.retrieve(action, question);
+              ModelResponse response =
+                  genkit.generate(
+                      GenerateOptions.builder()
+                          .model("googleai/gemini-2.5-flash")
+                          .system(RAG_SYSTEM_PROMPT)
+                          .prompt(question)
+                          .docs(docs)
+                          .config(GenerationConfig.builder().temperature(0.3).build())
+                          .build());
+              return response.getText();
+            });
+
+    logger.info(
+        "Genkit Milvus RAG Sample started. Flows: indexDocuments, retrieveDocuments, ragQuery");
+    logger.info("Dev UI: http://localhost:4000  |  Reflection: http://localhost:3100");
+
+    try {
+      jetty.start();
+    } catch (Exception e) {
+      logger.error("Failed to start Jetty server", e);
+      System.exit(1);
+    }
+  }
+
+  private static String getEnv(Dotenv dotenv, String name) {
+    String value = dotenv.get(name);
+    return (value != null && !value.isBlank()) ? value : System.getenv(name);
+  }
+
+  private static String getEnvOrDefault(Dotenv dotenv, String name, String defaultValue) {
+    String value = getEnv(dotenv, name);
+    return (value != null && !value.isBlank()) ? value : defaultValue;
+  }
+}
diff --git a/samples/milvus/src/main/resources/logback.xml b/samples/milvus/src/main/resources/logback.xml
new file mode 100644
index 000000000..dff2540b6
--- /dev/null
+++ b/samples/milvus/src/main/resources/logback.xml
@@ -0,0 +1,20 @@
+
+
+    
+        
+            %d{HH:mm:ss.SSS} %-5level %logger{24} - %msg%n
+        
+    
+
+    
+    
+    
+    
+    
+
+    
+
+    
+        
+    
+
diff --git a/samples/milvus/user.yaml b/samples/milvus/user.yaml
new file mode 100644
index 000000000..8d312694b
--- /dev/null
+++ b/samples/milvus/user.yaml
@@ -0,0 +1 @@
+# Extra config to override default milvus.yaml

From a0fee7851124959e9a41d905b7f4fd2b3bdc06ec Mon Sep 17 00:00:00 2001
From: xavidop 
Date: Mon, 6 Jul 2026 14:37:46 +0200
Subject: [PATCH 3/6] fix: remove user.yaml

---
 samples/milvus/user.yaml | 1 -
 1 file changed, 1 deletion(-)
 delete mode 100644 samples/milvus/user.yaml

diff --git a/samples/milvus/user.yaml b/samples/milvus/user.yaml
deleted file mode 100644
index 8d312694b..000000000
--- a/samples/milvus/user.yaml
+++ /dev/null
@@ -1 +0,0 @@
-# Extra config to override default milvus.yaml

From e99012c29278dea922c8ff46afb170ae4ff935ff Mon Sep 17 00:00:00 2001
From: xavidop 
Date: Mon, 6 Jul 2026 18:59:15 +0200
Subject: [PATCH 4/6] feat: update of the plugins

---
 docs/src/content/docs/plugins/anthropic.md    |  21 +-
 docs/src/content/docs/plugins/aws-bedrock.md  |  13 +
 docs/src/content/docs/plugins/cohere.md       |  17 +-
 docs/src/content/docs/plugins/deepseek.md     |   8 +-
 docs/src/content/docs/plugins/google-genai.md |  41 +-
 docs/src/content/docs/plugins/groq.md         |   4 +-
 docs/src/content/docs/plugins/mistral.md      |  23 +-
 docs/src/content/docs/plugins/openai.md       |  10 +-
 docs/src/content/docs/plugins/xai.md          |  11 +-
 plugins/anthropic/README.md                   |  14 +-
 .../plugins/anthropic/AnthropicModel.java     |  10 +-
 .../plugins/anthropic/AnthropicPlugin.java    |  19 +-
 .../anthropic/AnthropicPluginTest.java        |   3 +-
 plugins/aws-bedrock/README.md                 |  19 +-
 .../awsbedrock/AwsBedrockEmbedder.java        | 172 ++++++++
 .../plugins/awsbedrock/AwsBedrockModel.java   |  50 +--
 .../plugins/awsbedrock/AwsBedrockPlugin.java  |  56 ++-
 .../plugins/awsbedrock/AwsBedrockSigner.java  | 109 +++++
 .../awsbedrock/AwsBedrockPluginTest.java      |  18 +
 plugins/azure-foundry/README.md               |  20 +-
 .../azurefoundry/AzureFoundryPlugin.java      |  43 +-
 plugins/cohere/README.md                      |  10 +
 .../genkit/plugins/cohere/CoherePlugin.java   |  44 +-
 .../plugins/cohere/CoherePluginTest.java      |  20 +
 .../plugins/compatoai/CompatOAIEmbedder.java  | 178 ++++++++
 .../plugins/deepseek/DeepSeekPlugin.java      |   8 +-
 .../plugins/googlegenai/GeminiEmbedder.java   |   2 +-
 .../googlegenai/GoogleGenAIPlugin.java        |  83 ++--
 .../plugins/googlegenai/ImagenModel.java      |   2 +-
 .../genkit/plugins/googlegenai/OmniModel.java | 397 ++++++++++++++++++
 .../genkit/plugins/googlegenai/TtsModel.java  |  34 +-
 .../genkit/plugins/googlegenai/VeoModel.java  |  10 +-
 .../googlegenai/GoogleGenAIPluginTest.java    |  21 +
 .../plugins/googlegenai/TtsModelTest.java     |  52 +++
 plugins/groq/README.md                        |   5 +-
 .../genkit/plugins/groq/GroqPlugin.java       |   7 +-
 plugins/mistral/README.md                     |  23 +-
 .../genkit/plugins/mistral/MistralPlugin.java |  44 +-
 .../plugins/mistral/MistralPluginTest.java    |  20 +
 plugins/openai/README.md                      |  11 +-
 .../genkit/plugins/openai/OpenAIPlugin.java   |  26 +-
 .../plugins/openai/OpenAIPluginTest.java      |   2 +-
 plugins/xai/README.md                         |  11 +-
 .../google/genkit/plugins/xai/XAIPlugin.java  |  27 +-
 .../genkit/plugins/xai/XAIPluginTest.java     |   7 +-
 samples/anthropic/README.md                   |  14 +-
 .../genkit/samples/AzureFoundrySample.java    |   6 +-
 .../samples/firebase/FirestoreRAGSample.java  |   2 +-
 samples/google-genai/README.md                |   8 +-
 .../google/genkit/samples/GoogleGenAIApp.java |  91 +++-
 samples/groq/README.md                        |   9 +-
 .../com/google/genkit/samples/GroqSample.java |  12 +-
 samples/mistral/README.md                     |   3 -
 .../google/genkit/samples/MistralSample.java  |   2 +-
 samples/pinecone/README.md                    |   2 +-
 samples/xai/README.md                         |  35 +-
 .../com/google/genkit/samples/XAISample.java  |  12 +-
 57 files changed, 1627 insertions(+), 294 deletions(-)
 create mode 100644 plugins/aws-bedrock/src/main/java/com/google/genkit/plugins/awsbedrock/AwsBedrockEmbedder.java
 create mode 100644 plugins/aws-bedrock/src/main/java/com/google/genkit/plugins/awsbedrock/AwsBedrockSigner.java
 create mode 100644 plugins/compat-oai/src/main/java/com/google/genkit/plugins/compatoai/CompatOAIEmbedder.java
 create mode 100644 plugins/google-genai/src/main/java/com/google/genkit/plugins/googlegenai/OmniModel.java
 create mode 100644 plugins/google-genai/src/test/java/com/google/genkit/plugins/googlegenai/TtsModelTest.java

diff --git a/docs/src/content/docs/plugins/anthropic.md b/docs/src/content/docs/plugins/anthropic.md
index d7ce87b87..34dbdb67b 100644
--- a/docs/src/content/docs/plugins/anthropic.md
+++ b/docs/src/content/docs/plugins/anthropic.md
@@ -32,25 +32,26 @@ Genkit genkit = Genkit.builder()
 
 ModelResponse response = genkit.generate(
     GenerateOptions.builder()
-        .model("anthropic/claude-sonnet-4-5-20250929")
+        .model("anthropic/claude-sonnet-5")
         .prompt("Tell me about AI")
         .build());
 ```
 
 ## Available models
 
-### Claude 4.5 family
+### Claude 5 family
+- `anthropic/claude-fable-5`
+- `anthropic/claude-sonnet-5`
+
+### Claude 4 family
+- `anthropic/claude-opus-4-8`
+- `anthropic/claude-opus-4-7`
+- `anthropic/claude-opus-4-6`
+- `anthropic/claude-sonnet-4-6`
 - `anthropic/claude-opus-4-5-20251101`
 - `anthropic/claude-sonnet-4-5-20250929`
 - `anthropic/claude-haiku-4-5-20251001`
-
-### Claude 4 family
-- `anthropic/claude-4-*`
-
-### Claude 3 family
-- `anthropic/claude-3-opus-*`
-- `anthropic/claude-3-sonnet-*`
-- `anthropic/claude-3-haiku-*`
+- `anthropic/claude-opus-4-1-20250805`
 
 ## Features
 
diff --git a/docs/src/content/docs/plugins/aws-bedrock.md b/docs/src/content/docs/plugins/aws-bedrock.md
index b13d93b61..9dd4a8f5f 100644
--- a/docs/src/content/docs/plugins/aws-bedrock.md
+++ b/docs/src/content/docs/plugins/aws-bedrock.md
@@ -51,6 +51,19 @@ ModelResponse response = genkit.generate(
 - **OpenAI** — Via Bedrock marketplace
 - And many more...
 
+## Embeddings
+
+Bedrock embedding models (Amazon Titan and Cohere Embed) are available via the `InvokeModel` API:
+
+- `aws-bedrock/amazon.titan-embed-text-v2:0`
+- `aws-bedrock/cohere.embed-english-v3`
+- `aws-bedrock/cohere.embed-multilingual-v3`
+
+```java
+EmbedResponse response =
+    genkit.embed("aws-bedrock/amazon.titan-embed-text-v2:0", documents);
+```
+
 ## Features
 
 - Multi-provider model access
diff --git a/docs/src/content/docs/plugins/cohere.md b/docs/src/content/docs/plugins/cohere.md
index 00f222be1..8a62e786a 100644
--- a/docs/src/content/docs/plugins/cohere.md
+++ b/docs/src/content/docs/plugins/cohere.md
@@ -37,14 +37,29 @@ ModelResponse response = genkit.generate(
 
 ## Available models
 
+- `cohere/command-a-plus-05-2026` — flagship (Mixture-of-Experts, text + vision)
+- `cohere/command-a-reasoning-08-2025`
+- `cohere/command-a-vision-07-2025`
 - `cohere/command-a-03-2025`
 - `cohere/command-r7b-12-2024`
 - `cohere/command-r-08-2024`
 - `cohere/command-r-plus-08-2024`
 
+## Embeddings
+
+Cohere embedding models are available through the OpenAI-compatible endpoint:
+
+- `cohere/embed-v4.0`
+- `cohere/embed-multilingual-v3.0`
+- `cohere/embed-english-v3.0`
+
+```java
+EmbedResponse response = genkit.embed("cohere/embed-v4.0", documents);
+```
+
 ## Features
 
-- Text generation, tool calling, RAG support, SSE streaming
+- Text generation, tool calling, RAG support, SSE streaming, embeddings
 
 ## Sample
 
diff --git a/docs/src/content/docs/plugins/deepseek.md b/docs/src/content/docs/plugins/deepseek.md
index a19a0b247..133f268c5 100644
--- a/docs/src/content/docs/plugins/deepseek.md
+++ b/docs/src/content/docs/plugins/deepseek.md
@@ -30,15 +30,17 @@ Genkit genkit = Genkit.builder()
 
 ModelResponse response = genkit.generate(
     GenerateOptions.builder()
-        .model("deepseek/deepseek-chat")
+        .model("deepseek/deepseek-v4-pro")
         .prompt("Tell me about AI")
         .build());
 ```
 
 ## Available models
 
-- `deepseek/deepseek-chat` — General chat model
-- `deepseek/deepseek-reasoner` — Reasoning model
+- `deepseek/deepseek-v4-pro` — Flagship V4 model
+- `deepseek/deepseek-v4-flash` — Fast, economical V4 model
+- `deepseek/deepseek-chat` — Legacy alias (deprecated, retires 2026-07-24)
+- `deepseek/deepseek-reasoner` — Legacy reasoning alias (deprecated, retires 2026-07-24)
 
 ## Features
 
diff --git a/docs/src/content/docs/plugins/google-genai.md b/docs/src/content/docs/plugins/google-genai.md
index f3a5224f3..28cd23823 100644
--- a/docs/src/content/docs/plugins/google-genai.md
+++ b/docs/src/content/docs/plugins/google-genai.md
@@ -85,7 +85,7 @@ Map embedOptions = Map.of(
 
 ## Text-to-Speech (TTS)
 
-Generate natural-sounding speech from text using Gemini TTS models:
+Generate natural-sounding speech from text using Gemini TTS models (`gemini-3.1-flash-tts-preview`, `gemini-2.5-flash-preview-tts`, `gemini-2.5-pro-preview-tts`):
 
 ```java
 Map ttsOptions = Map.of("voiceName", "Zephyr");
@@ -105,6 +105,12 @@ String audioDataUrl = response.getMessage().getParts().get(0).getMedia().getUrl(
 // "data:audio/wav;base64,..."
 ```
 
+The plugin automatically frames your prompt with a synthesis preamble so the TTS model treats it as a transcript to voice. Without this, TTS models reject "vague" prompts with a `400` error (`Model tried to generate text, but it should only be used for TTS`). To supply your own framing (e.g. `"Say cheerfully: ..."`), set a `ttsInstruction` custom option — a preamble string, or an empty string to send the prompt verbatim:
+
+```java
+Map ttsOptions = Map.of("voiceName", "Zephyr", "ttsInstruction", "");
+```
+
 ### Saving audio to a file
 
 ```java
@@ -131,7 +137,7 @@ GenerationConfig config = GenerationConfig.builder()
 
 ModelResponse response = genkit.generate(
     GenerateOptions.builder()
-        .model("googleai/veo-3.0-generate-001")
+        .model("googleai/veo-3.1-generate-preview")
         .prompt("A serene Japanese garden with cherry blossoms falling")
         .config(config)
         .build());
@@ -167,6 +173,37 @@ ModelResponse response = genkit.generate(
 ```
 
 
+## Video generation and editing (Gemini Omni)
+
+Gemini Omni (`googleai/gemini-omni-flash-preview`) generates and iteratively edits video through the Gemini Interactions API. It supports **conversational editing** — each turn can build on the previous result while preserving elements you did not mention.
+
+```java
+// First turn — generate a video
+ModelResponse first = genkit.generate(
+    GenerateOptions.builder()
+        .model("googleai/gemini-omni-flash-preview")
+        .prompt("A marble rolling down a chain-reaction track")
+        .build());
+
+// The returned interaction id lets you continue editing
+String interactionId = (String) first.getCustom().get("interactionId");
+
+// Follow-up turn — edit the previous result
+ModelResponse edited = genkit.generate(
+    GenerateOptions.builder()
+        .model("googleai/gemini-omni-flash-preview")
+        .prompt("Brighten the background and add a slow push-in on the logo")
+        .config(GenerationConfig.builder()
+            .custom(Map.of("previousInteractionId", interactionId))
+            .build())
+        .build());
+
+// The generated video is returned as a media part (data URL by default)
+String videoUrl = edited.getMessage().getParts().get(0).getMedia().getUrl();
+```
+
+Config options: `previousInteractionId` (continue/edit a prior interaction), `aspectRatio` (e.g. `"16:9"`), `duration` (e.g. `"10s"`), `delivery` (`"inline"` (default, base64) | `"uri"`), `task` (`"text_to_video"` | `"image_to_video"`), `thinkingLevel`, and `maxOutputTokens`. Only the Gemini Developer API (API key) is supported — not Vertex AI. The Interactions API is in preview.
+
 ## Sample
 
 See the [google-genai sample](https://github.com/genkit-ai/genkit-java/tree/main/samples/google-genai) for complete examples of text generation, tool calling, embeddings, image generation, TTS, and video generation.
diff --git a/docs/src/content/docs/plugins/groq.md b/docs/src/content/docs/plugins/groq.md
index 8088038f2..101963643 100644
--- a/docs/src/content/docs/plugins/groq.md
+++ b/docs/src/content/docs/plugins/groq.md
@@ -41,8 +41,8 @@ ModelResponse response = genkit.generate(
 |-------|-------|
 | `groq/llama-3.1-8b-instant` | ~1200 tokens/sec |
 | `groq/llama-3.3-70b-versatile` | ~560 tokens/sec |
-| `groq/openai/gpt-oss` models | Varies |
-| `groq/meta-llama/llama-guard-4` | Content moderation |
+| `groq/openai/gpt-oss-120b`, `groq/openai/gpt-oss-20b` | Varies |
+| `groq/compound`, `groq/compound-mini` | Agentic systems (web search + code execution) |
 
 ## Features
 
diff --git a/docs/src/content/docs/plugins/mistral.md b/docs/src/content/docs/plugins/mistral.md
index 79ca024fa..df1f03485 100644
--- a/docs/src/content/docs/plugins/mistral.md
+++ b/docs/src/content/docs/plugins/mistral.md
@@ -30,7 +30,7 @@ Genkit genkit = Genkit.builder()
 
 ModelResponse response = genkit.generate(
     GenerateOptions.builder()
-        .model("mistral/mistral-large-3-25-12")
+        .model("mistral/mistral-large-2512")
         .prompt("Tell me about AI")
         .build());
 ```
@@ -39,15 +39,24 @@ ModelResponse response = genkit.generate(
 
 | Model | Context |
 |-------|---------|
-| `mistral/mistral-large-3-25-12` | 128K |
-| `mistral/mistral-medium-3-1-25-08` | 128K |
-| `mistral/mistral-small-*` | 128K |
-| `mistral/ministral-*` | 128K |
-| `mistral/codestral-25-08` | 256K |
+| `mistral/mistral-large-2512` | 128K |
+| `mistral/mistral-medium-2604` | 128K |
+| `mistral/mistral-small-2603` | 128K |
+| `mistral/ministral-3b-2512`, `mistral/ministral-8b-2512`, `mistral/ministral-14b-2512` | 128K |
+| `mistral/codestral-2508` | 256K |
+
+## Embeddings
+
+- `mistral/mistral-embed`
+- `mistral/codestral-embed`
+
+```java
+EmbedResponse response = genkit.embed("mistral/mistral-embed", documents);
+```
 
 ## Features
 
-- Text generation, streaming, tool calling, RAG
+- Text generation, streaming, tool calling, RAG, embeddings
 
 ## Sample
 
diff --git a/docs/src/content/docs/plugins/openai.md b/docs/src/content/docs/plugins/openai.md
index 3c82ecd17..531471987 100644
--- a/docs/src/content/docs/plugins/openai.md
+++ b/docs/src/content/docs/plugins/openai.md
@@ -39,11 +39,13 @@ ModelResponse response = genkit.generate(
 
 ## Available models
 
-- `openai/gpt-4o` — Most capable model
+- `openai/gpt-5.5` — Most capable flagship (also `gpt-5.5-pro`)
+- `openai/gpt-5.4` — High-performance general model (also `-mini`, `-nano`, `-pro`)
+- `openai/gpt-5` — GPT-5 base (also `gpt-5-mini`, `gpt-5-nano`, `gpt-5-pro`)
+- `openai/gpt-4.1` — Smartest non-reasoning model (also `gpt-4.1-mini`)
+- `openai/gpt-4o` — Multimodal model
 - `openai/gpt-4o-mini` — Fast and cost-effective
-- `openai/gpt-4-turbo` — Previous generation flagship
-- `openai/o1-preview` — Reasoning model
-- `openai/o1-mini` — Fast reasoning model
+- `openai/o3` — Reasoning model (also `o3-pro`)
 
 ## Embeddings
 
diff --git a/docs/src/content/docs/plugins/xai.md b/docs/src/content/docs/plugins/xai.md
index 440202d3d..ee2e4cc3c 100644
--- a/docs/src/content/docs/plugins/xai.md
+++ b/docs/src/content/docs/plugins/xai.md
@@ -32,7 +32,7 @@ Genkit genkit = Genkit.builder()
 
 ModelResponse response = genkit.generate(
     GenerateOptions.builder()
-        .model("xai/grok-3")
+        .model("xai/grok-4.3")
         .prompt("Tell me about AI")
         .build());
 ```
@@ -41,10 +41,11 @@ ModelResponse response = genkit.generate(
 
 | Model | Context Window |
 |-------|---------------|
-| `xai/grok-4` | Up to 2M tokens |
-| `xai/grok-4-1-fast` | 131K tokens |
-| `xai/grok-3` | 131K tokens |
-| `xai/grok-3-mini` | 131K tokens |
+| `xai/grok-4.3` | Up to 2M tokens |
+| `xai/grok-4.20-0309-reasoning` | Up to 2M tokens |
+| `xai/grok-4.20-0309-non-reasoning` | Up to 2M tokens |
+| `xai/grok-4.20-multi-agent-0309` | Up to 2M tokens |
+| `xai/grok-build-0.1` | 256K tokens (agentic coding) |
 
 ## Features
 
diff --git a/plugins/anthropic/README.md b/plugins/anthropic/README.md
index 047f9e42a..b4527090a 100644
--- a/plugins/anthropic/README.md
+++ b/plugins/anthropic/README.md
@@ -79,17 +79,19 @@ AnthropicPlugin plugin = new AnthropicPlugin(
 - `anthropic/claude-sonnet-4-5-20250929` - Balanced performance (recommended)
 - `anthropic/claude-haiku-4-5-20251001` - Fast and efficient
 
+### Claude 5 Family
+- `anthropic/claude-fable-5` - Most capable widely released model
+- `anthropic/claude-sonnet-5` - Balanced speed and intelligence
+
 ### Claude 4 Family
+- `anthropic/claude-opus-4-8` - Flagship Opus
+- `anthropic/claude-opus-4-7` - Claude Opus 4.7
+- `anthropic/claude-opus-4-6` - Claude Opus 4.6
+- `anthropic/claude-sonnet-4-6` - Claude Sonnet 4.6
 - `anthropic/claude-opus-4-1-20250805` - Claude Opus 4.1
 - `anthropic/claude-opus-4-20250514` - Claude Opus 4
 - `anthropic/claude-sonnet-4-20250514` - Claude Sonnet 4
 
-### Claude 3 Family
-- `anthropic/claude-3-7-sonnet-20250219` - Claude Sonnet 3.7
-- `anthropic/claude-3-5-haiku-20241022` - Claude Haiku 3.5
-- `anthropic/claude-3-opus-20240229` - Claude Opus 3
-- `anthropic/claude-3-haiku-20240307` - Claude Haiku 3
-
 ## Using Custom Models
 
 If you need to use a model not in the default list (e.g., a newer model release), register it using `customModel()`:
diff --git a/plugins/anthropic/src/main/java/com/google/genkit/plugins/anthropic/AnthropicModel.java b/plugins/anthropic/src/main/java/com/google/genkit/plugins/anthropic/AnthropicModel.java
index aa61fb677..f727514f9 100644
--- a/plugins/anthropic/src/main/java/com/google/genkit/plugins/anthropic/AnthropicModel.java
+++ b/plugins/anthropic/src/main/java/com/google/genkit/plugins/anthropic/AnthropicModel.java
@@ -39,8 +39,8 @@
 /**
  * Anthropic Claude model implementation for Genkit.
  *
- * 
Supports Claude 3.5, Claude 3, and Claude 2 model families with both synchronous and streaming
- * generation.
+ * 
Supports the Claude 5, Claude 4, and Claude 3.5 model families with both synchronous and
+ * streaming generation.
  */
 public class AnthropicModel implements Model {
 
@@ -78,9 +78,9 @@ private ModelInfo createModelInfo() {
 
     ModelInfo.ModelCapabilities caps = new ModelInfo.ModelCapabilities();
     caps.setMultiturn(true);
-    // Claude 3+ models support vision
-    caps.setMedia(modelName.contains("claude-3"));
-    caps.setTools(modelName.contains("claude-3"));
+    // All supported Claude models (3.5, 4.x, and 5 families) support vision and tool use.
+    caps.setMedia(true);
+    caps.setTools(true);
     caps.setSystemRole(true);
     caps.setOutput(Set.of("text", "json"));
     info.setSupports(caps);
diff --git a/plugins/anthropic/src/main/java/com/google/genkit/plugins/anthropic/AnthropicPlugin.java b/plugins/anthropic/src/main/java/com/google/genkit/plugins/anthropic/AnthropicPlugin.java
index 44d20f027..bbfcc0bfe 100644
--- a/plugins/anthropic/src/main/java/com/google/genkit/plugins/anthropic/AnthropicPlugin.java
+++ b/plugins/anthropic/src/main/java/com/google/genkit/plugins/anthropic/AnthropicPlugin.java
@@ -29,8 +29,8 @@
 /**
  * AnthropicPlugin provides Anthropic Claude model integrations for Genkit.
  *
- * 
This plugin registers Claude models (Claude 3.5, Claude 3, Claude 2) as Genkit actions for
- * text generation with support for streaming.
+ * 
This plugin registers Claude models (Claude 5, Claude 4, and Claude 3.5 families) as Genkit
+ * actions for text generation with support for streaming.
  */
 public class AnthropicPlugin implements Plugin {
 
@@ -39,6 +39,14 @@ public class AnthropicPlugin implements Plugin {
   /** Supported Claude models. */
   public static final List SUPPORTED_MODELS =
       Arrays.asList(
+          // Claude 5 family
+          "claude-fable-5",
+          "claude-sonnet-5",
+          // Claude 4.8 / 4.7 / 4.6
+          "claude-opus-4-8",
+          "claude-opus-4-7",
+          "claude-opus-4-6",
+          "claude-sonnet-4-6",
           // Claude 4.5 family
           "claude-opus-4-5-20251101",
           "claude-sonnet-4-5-20250929",
@@ -46,12 +54,7 @@ public class AnthropicPlugin implements Plugin {
           // Claude 4.x family
           "claude-opus-4-1-20250805",
           "claude-opus-4-20250514",
-          "claude-sonnet-4-20250514",
-          // Claude 3.x family
-          "claude-3-7-sonnet-20250219",
-          "claude-3-5-haiku-20241022",
-          "claude-3-opus-20240229",
-          "claude-3-haiku-20240307");
+          "claude-sonnet-4-20250514");
 
   private final AnthropicPluginOptions options;
   private final List customModels = new ArrayList<>();
diff --git a/plugins/anthropic/src/test/java/com/google/genkit/plugins/anthropic/AnthropicPluginTest.java b/plugins/anthropic/src/test/java/com/google/genkit/plugins/anthropic/AnthropicPluginTest.java
index fde246912..39aab3804 100644
--- a/plugins/anthropic/src/test/java/com/google/genkit/plugins/anthropic/AnthropicPluginTest.java
+++ b/plugins/anthropic/src/test/java/com/google/genkit/plugins/anthropic/AnthropicPluginTest.java
@@ -82,7 +82,8 @@ void testSupportedModels() {
     assertNotNull(AnthropicPlugin.SUPPORTED_MODELS);
     assertTrue(AnthropicPlugin.SUPPORTED_MODELS.contains("claude-opus-4-5-20251101"));
     assertTrue(AnthropicPlugin.SUPPORTED_MODELS.contains("claude-sonnet-4-5-20250929"));
-    assertTrue(AnthropicPlugin.SUPPORTED_MODELS.contains("claude-3-opus-20240229"));
+    assertTrue(AnthropicPlugin.SUPPORTED_MODELS.contains("claude-opus-4-8"));
+    assertTrue(AnthropicPlugin.SUPPORTED_MODELS.contains("claude-sonnet-5"));
   }
 
   @Test
diff --git a/plugins/aws-bedrock/README.md b/plugins/aws-bedrock/README.md
index 07cc80451..6334aeb42 100644
--- a/plugins/aws-bedrock/README.md
+++ b/plugins/aws-bedrock/README.md
@@ -41,17 +41,16 @@ System.out.println(response.getText());
 
 ### Amazon Models (ON_DEMAND)
 - **Nova**: `amazon.nova-pro-v1:0`, `amazon.nova-lite-v1:0`, `amazon.nova-micro-v1:0`, `amazon.nova-premier-v1:0`, `amazon.nova-2-lite-v1:0`, `amazon.nova-sonic-v1:0`, `amazon.nova-2-sonic-v1:0`
-- **Titan**: `amazon.titan-tg1-large`, `amazon.titan-text-express-v1`
 
 ### Anthropic Models (INFERENCE_PROFILE required for Claude 4.x and 3.5+)
+- **Claude 5 / 4.6+**: `anthropic.claude-sonnet-5`, `anthropic.claude-fable-5`, `anthropic.claude-opus-4-8`, `anthropic.claude-opus-4-7`, `anthropic.claude-opus-4-6-v1`, `anthropic.claude-sonnet-4-6`
 - **Claude 4**: `anthropic.claude-sonnet-4-20250514-v1:0`, `anthropic.claude-sonnet-4-5-20250929-v1:0`, `anthropic.claude-haiku-4-5-20251001-v1:0`, `anthropic.claude-opus-4-1-20250805-v1:0`, `anthropic.claude-opus-4-5-20251101-v1:0`
 - **Claude 3.x**: `anthropic.claude-3-7-sonnet-20250219-v1:0`, `anthropic.claude-3-5-sonnet-20241022-v2:0`, `anthropic.claude-3-5-sonnet-20240620-v1:0`, `anthropic.claude-3-5-haiku-20241022-v1:0`
-- **Claude 3** (ON_DEMAND): `anthropic.claude-3-opus-20240229-v1:0`, `anthropic.claude-3-sonnet-20240229-v1:0`, `anthropic.claude-3-haiku-20240307-v1:0`
+- **Claude 3** (ON_DEMAND): `anthropic.claude-3-sonnet-20240229-v1:0`, `anthropic.claude-3-haiku-20240307-v1:0`
 
 ### Meta Llama Models (INFERENCE_PROFILE)
 - **Llama 4**: `meta.llama4-scout-17b-instruct-v1:0`, `meta.llama4-maverick-17b-instruct-v1:0`
 - **Llama 3.3**: `meta.llama3-3-70b-instruct-v1:0`
-- **Llama 3.2**: `meta.llama3-2-90b-instruct-v1:0`, `meta.llama3-2-11b-instruct-v1:0`, `meta.llama3-2-3b-instruct-v1:0`, `meta.llama3-2-1b-instruct-v1:0`
 - **Llama 3.1**: `meta.llama3-1-70b-instruct-v1:0`, `meta.llama3-1-8b-instruct-v1:0`
 - **Llama 3** (ON_DEMAND): `meta.llama3-70b-instruct-v1:0`, `meta.llama3-8b-instruct-v1:0`
 
@@ -60,11 +59,12 @@ System.out.println(response.getText());
 - **Mistral Large** (ON_DEMAND): `mistral.mistral-large-3-675b-instruct`, `mistral.mistral-large-2402-v1:0`
 - **Mistral Medium**: `mistral.magistral-small-2509`, `mistral.mistral-small-2402-v1:0`
 - **Ministral**: `mistral.ministral-3-14b-instruct`, `mistral.ministral-3-8b-instruct`, `mistral.ministral-3-3b-instruct`
+- **Devstral** (INFERENCE_PROFILE): `mistral.devstral-2-123b`
 - **Mixtral**: `mistral.mixtral-8x7b-instruct-v0:1`, `mistral.mistral-7b-instruct-v0:2`
 - **Voxtral**: `mistral.voxtral-mini-3b-2507`, `mistral.voxtral-small-24b-2507`
 
 ### Other Providers (ON_DEMAND)
-- **DeepSeek**: `deepseek.r1-v1:0` (INFERENCE_PROFILE)
+- **DeepSeek**: `deepseek.v3.2`, `deepseek.v3-v1:0`, `deepseek.r1-v1:0` (INFERENCE_PROFILE)
 - **Cohere**: `cohere.command-r-plus-v1:0`, `cohere.command-r-v1:0`
 - **AI21 Labs**: `ai21.jamba-1-5-large-v1:0`, `ai21.jamba-1-5-mini-v1:0`
 - **Google Gemma**: `google.gemma-3-27b-it`, `google.gemma-3-12b-it`, `google.gemma-3-4b-it`
@@ -72,10 +72,17 @@ System.out.println(response.getText());
 - **NVIDIA Nemotron**: `nvidia.nemotron-nano-12b-v2`, `nvidia.nemotron-nano-9b-v2`, `nvidia.nemotron-nano-3-30b`
 - **OpenAI**: `openai.gpt-oss-120b-1:0`, `openai.gpt-oss-20b-1:0`, `openai.gpt-oss-safeguard-120b`, `openai.gpt-oss-safeguard-20b`
 - **Writer** (INFERENCE_PROFILE): `writer.palmyra-x5-v1:0`, `writer.palmyra-x4-v1:0`
-- **MiniMax**: `minimax.minimax-m2`
-- **Moonshot**: `moonshot.kimi-k2-thinking`
+- **MiniMax**: `minimax.minimax-m2.5`, `minimax.minimax-m2.1`, `minimax.minimax-m2`
+- **Moonshot**: `moonshotai.kimi-k2.5`, `moonshot.kimi-k2-thinking`
+- **Z.AI**: `zai.glm-5`, `zai.glm-4.7`, `zai.glm-4.7-flash`
 - **TwelveLabs**: `twelvelabs.pegasus-1-2-v1:0`
 
+### Embedding Models (InvokeModel)
+- **Amazon Titan**: `amazon.titan-embed-text-v2:0`
+- **Cohere Embed**: `cohere.embed-english-v3`, `cohere.embed-multilingual-v3`
+
+Register additional embedding models with `customEmbeddingModel(...)`.
+
 **Note**: Models marked with "INFERENCE_PROFILE" require using an AWS Bedrock inference profile ARN instead of the model ID directly. See [AWS Bedrock Inference Profiles](https://docs.aws.amazon.com/bedrock/latest/userguide/cross-region-inference.html) for details.
 
 ## Configuration
diff --git a/plugins/aws-bedrock/src/main/java/com/google/genkit/plugins/awsbedrock/AwsBedrockEmbedder.java b/plugins/aws-bedrock/src/main/java/com/google/genkit/plugins/awsbedrock/AwsBedrockEmbedder.java
new file mode 100644
index 000000000..4b884f637
--- /dev/null
+++ b/plugins/aws-bedrock/src/main/java/com/google/genkit/plugins/awsbedrock/AwsBedrockEmbedder.java
@@ -0,0 +1,172 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package com.google.genkit.plugins.awsbedrock;
+
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.fasterxml.jackson.databind.node.ObjectNode;
+import com.google.genkit.ai.Document;
+import com.google.genkit.ai.EmbedRequest;
+import com.google.genkit.ai.EmbedResponse;
+import com.google.genkit.ai.Embedder;
+import com.google.genkit.ai.EmbedderInfo;
+import com.google.genkit.core.ActionContext;
+import com.google.genkit.core.GenkitException;
+import java.io.IOException;
+import java.util.ArrayList;
+import java.util.List;
+import java.util.concurrent.TimeUnit;
+import okhttp3.OkHttpClient;
+import okhttp3.Request;
+import okhttp3.Response;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+/**
+ * AWS Bedrock embedder implementation for Genkit.
+ *
+ * 
Uses the AWS Bedrock {@code InvokeModel} API (via HTTP with AWS SigV4 signing) to generate
+ * embeddings. Handles the differing request/response shapes of Amazon Titan Text Embeddings and
+ * Cohere Embed models.
+ */
+public class AwsBedrockEmbedder extends Embedder {
+
+  private static final Logger logger = LoggerFactory.getLogger(AwsBedrockEmbedder.class);
+
+  private final String modelId;
+  private final AwsBedrockPluginOptions options;
+  private final OkHttpClient client;
+  private final ObjectMapper objectMapper;
+
+  /**
+   * Creates a new AwsBedrockEmbedder.
+   *
+   * @param modelId the Bedrock embedding model ID (e.g., "amazon.titan-embed-text-v2:0")
+   * @param options the plugin options
+   */
+  public AwsBedrockEmbedder(String modelId, AwsBedrockPluginOptions options) {
+    super(
+        "aws-bedrock/" + modelId,
+        createEmbedderInfo(modelId),
+        (ctx, req) -> {
+          throw new GenkitException("Handler not initialized");
+        });
+    this.modelId = modelId;
+    this.options = options;
+    this.objectMapper = new ObjectMapper();
+    this.client =
+        new OkHttpClient.Builder()
+            .connectTimeout(120, TimeUnit.SECONDS)
+            .readTimeout(120, TimeUnit.SECONDS)
+            .writeTimeout(120, TimeUnit.SECONDS)
+            .build();
+  }
+
+  private static EmbedderInfo createEmbedderInfo(String modelId) {
+    EmbedderInfo info = new EmbedderInfo();
+    info.setLabel("AWS Bedrock " + modelId);
+    if (modelId.contains("titan-embed-text-v2")) {
+      info.setDimensions(1024);
+    } else if (modelId.contains("titan-embed-text-v1")) {
+      info.setDimensions(1536);
+    } else if (modelId.startsWith("cohere.embed")) {
+      info.setDimensions(1024);
+    }
+    return info;
+  }
+
+  private boolean isCohere() {
+    return modelId.startsWith("cohere.embed") || modelId.contains("cohere.embed");
+  }
+
+  @Override
+  public EmbedResponse run(ActionContext context, EmbedRequest request) {
+    if (request == null) {
+      throw new GenkitException(
+          "Embed request is required. Please provide an input with documents to embed.");
+    }
+    if (request.getDocuments() == null || request.getDocuments().isEmpty()) {
+      throw new GenkitException("Embed request must contain at least one document to embed.");
+    }
+    try {
+      List embeddings = new ArrayList<>();
+      for (Document doc : request.getDocuments()) {
+        String text = doc.text();
+        if (text == null || text.isEmpty()) {
+          logger.warn("Document has empty text, skipping");
+          continue;
+        }
+        embeddings.add(embedOne(text));
+      }
+      return new EmbedResponse(embeddings);
+    } catch (IOException e) {
+      throw new GenkitException("AWS Bedrock Embedding API call failed", e);
+    }
+  }
+
+  private EmbedResponse.Embedding embedOne(String text) throws IOException {
+    ObjectNode body = objectMapper.createObjectNode();
+    if (isCohere()) {
+      body.putArray("texts").add(text);
+      body.put("input_type", "search_document");
+    } else {
+      // Amazon Titan Text Embeddings
+      body.put("inputText", text);
+    }
+
+    String path = String.format("/model/%s/invoke", modelId);
+    String host = AwsBedrockSigner.runtimeHost(options);
+    Request httpRequest = AwsBedrockSigner.signRequest(options, host, path, body.toString());
+
+    try (Response response = client.newCall(httpRequest).execute()) {
+      if (!response.isSuccessful()) {
+        String errorBody = response.body() != null ? response.body().string() : "No error body";
+        throw new GenkitException(
+            "AWS Bedrock Embedding API error: " + response.code() + " - " + errorBody);
+      }
+      return parseEmbedding(objectMapper.readTree(response.body().string()));
+    }
+  }
+
+  private EmbedResponse.Embedding parseEmbedding(JsonNode root) {
+    // Titan: {"embedding": [...]}. Cohere: {"embeddings": [[...]]} or {"embeddings": {"float":
+    // [[...]]}}.
+    JsonNode vector = root.get("embedding");
+    if (vector == null) {
+      JsonNode embeddings = root.get("embeddings");
+      if (embeddings != null && embeddings.isArray() && embeddings.size() > 0) {
+        vector = embeddings.get(0);
+      } else if (embeddings != null && embeddings.has("float")) {
+        JsonNode floats = embeddings.get("float");
+        if (floats.isArray() && floats.size() > 0) {
+          vector = floats.get(0);
+        }
+      }
+    }
+    if (vector == null || !vector.isArray()) {
+      throw new GenkitException(
+          "AWS Bedrock embedding response did not contain an embedding vector");
+    }
+    float[] values = new float[vector.size()];
+    for (int i = 0; i < vector.size(); i++) {
+      values[i] = (float) vector.get(i).asDouble();
+    }
+    return new EmbedResponse.Embedding(values);
+  }
+}
diff --git a/plugins/aws-bedrock/src/main/java/com/google/genkit/plugins/awsbedrock/AwsBedrockModel.java b/plugins/aws-bedrock/src/main/java/com/google/genkit/plugins/awsbedrock/AwsBedrockModel.java
index 5d151b161..2bfa8a639 100644
--- a/plugins/aws-bedrock/src/main/java/com/google/genkit/plugins/awsbedrock/AwsBedrockModel.java
+++ b/plugins/aws-bedrock/src/main/java/com/google/genkit/plugins/awsbedrock/AwsBedrockModel.java
@@ -33,11 +33,6 @@
 import okhttp3.*;
 import org.slf4j.Logger;
 import org.slf4j.LoggerFactory;
-import software.amazon.awssdk.auth.credentials.AwsCredentials;
-import software.amazon.awssdk.auth.signer.Aws4Signer;
-import software.amazon.awssdk.auth.signer.params.Aws4SignerParams;
-import software.amazon.awssdk.http.SdkHttpFullRequest;
-import software.amazon.awssdk.http.SdkHttpMethod;
 
 /**
  * AWS Bedrock model implementation for Genkit.
@@ -170,50 +165,7 @@ private ModelResponse callBedrock(
   }
 
   private Request signRequest(String host, String path, String body) {
-    try {
-      AwsCredentials credentials = options.getCredentialsProvider().resolveCredentials();
-
-      // Build URI and let it handle encoding properly
-      java.net.URI uri = java.net.URI.create(String.format("https://%s%s", host, path));
-
-      SdkHttpFullRequest httpRequest =
-          SdkHttpFullRequest.builder()
-              .uri(uri)
-              .method(SdkHttpMethod.POST)
-              .putHeader("Content-Type", "application/json; charset=utf-8")
-              .contentStreamProvider(
-                  () ->
-                      new java.io.ByteArrayInputStream(
-                          body.getBytes(java.nio.charset.StandardCharsets.UTF_8)))
-              .build();
-
-      Aws4Signer signer = Aws4Signer.create();
-      Aws4SignerParams signerParams =
-          Aws4SignerParams.builder()
-              .awsCredentials(credentials)
-              .signingName("bedrock")
-              .signingRegion(options.getRegion())
-              .build();
-
-      SdkHttpFullRequest signedRequest = signer.sign(httpRequest, signerParams);
-
-      // Use the signed request's URI directly
-      Request.Builder okHttpRequestBuilder =
-          new Request.Builder()
-              .url(signedRequest.getUri().toURL())
-              .post(RequestBody.create(body, JSON_MEDIA_TYPE));
-
-      signedRequest
-          .headers()
-          .forEach(
-              (key, values) -> {
-                values.forEach(value -> okHttpRequestBuilder.addHeader(key, value));
-              });
-
-      return okHttpRequestBuilder.build();
-    } catch (Exception e) {
-      throw new GenkitException("Failed to sign AWS request", e);
-    }
+    return AwsBedrockSigner.signRequest(options, host, path, body);
   }
 
   private ObjectNode buildRequestBody(ModelRequest request, boolean stream) {
diff --git a/plugins/aws-bedrock/src/main/java/com/google/genkit/plugins/awsbedrock/AwsBedrockPlugin.java b/plugins/aws-bedrock/src/main/java/com/google/genkit/plugins/awsbedrock/AwsBedrockPlugin.java
index fa3f0a066..e19a1ec64 100644
--- a/plugins/aws-bedrock/src/main/java/com/google/genkit/plugins/awsbedrock/AwsBedrockPlugin.java
+++ b/plugins/aws-bedrock/src/main/java/com/google/genkit/plugins/awsbedrock/AwsBedrockPlugin.java
@@ -54,8 +54,16 @@ public class AwsBedrockPlugin implements Plugin {
 
   private static final Logger logger = LoggerFactory.getLogger(AwsBedrockPlugin.class);
 
+  /** Supported AWS Bedrock embedding models (Amazon Titan and Cohere Embed). */
+  public static final List SUPPORTED_EMBEDDING_MODELS =
+      Arrays.asList(
+          "amazon.titan-embed-text-v2:0",
+          "cohere.embed-english-v3",
+          "cohere.embed-multilingual-v3");
+
   private final AwsBedrockPluginOptions options;
   private final List customModels = new ArrayList<>();
+  private final List customEmbeddingModels = new ArrayList<>();
 
   /** Supported AWS Bedrock models with ON_DEMAND or INFERENCE_PROFILE support. */
   public static final List SUPPORTED_MODELS =
@@ -68,9 +76,6 @@ public class AwsBedrockPlugin implements Plugin {
           "amazon.nova-2-lite-v1:0",
           "amazon.nova-sonic-v1:0",
           "amazon.nova-2-sonic-v1:0",
-          // Amazon Titan models
-          "amazon.titan-tg1-large",
-          "amazon.titan-text-express-v1",
           // Anthropic Claude 4 models (INFERENCE_PROFILE)
           "anthropic.claude-sonnet-4-20250514-v1:0",
           "anthropic.claude-sonnet-4-5-20250929-v1:0",
@@ -91,7 +96,6 @@ public class AwsBedrockPlugin implements Plugin {
           "anthropic.claude-3-5-sonnet-20241022-v2:0",
           "anthropic.claude-3-5-sonnet-20240620-v1:0",
           "anthropic.claude-3-5-haiku-20241022-v1:0",
-          "anthropic.claude-3-opus-20240229-v1:0",
           "anthropic.claude-3-sonnet-20240229-v1:0",
           "anthropic.claude-3-haiku-20240307-v1:0",
           // AI21 models
@@ -101,10 +105,6 @@ public class AwsBedrockPlugin implements Plugin {
           "meta.llama4-scout-17b-instruct-v1:0",
           "meta.llama4-maverick-17b-instruct-v1:0",
           "meta.llama3-3-70b-instruct-v1:0",
-          "meta.llama3-2-90b-instruct-v1:0",
-          "meta.llama3-2-11b-instruct-v1:0",
-          "meta.llama3-2-3b-instruct-v1:0",
-          "meta.llama3-2-1b-instruct-v1:0",
           "meta.llama3-1-70b-instruct-v1:0",
           "meta.llama3-1-8b-instruct-v1:0",
           "meta.llama3-70b-instruct-v1:0",
@@ -115,6 +115,7 @@ public class AwsBedrockPlugin implements Plugin {
           // Mistral models
           "mistral.pixtral-large-2502-v1:0",
           "mistral.mistral-large-3-675b-instruct",
+          "mistral.devstral-2-123b",
           "mistral.magistral-small-2509",
           "mistral.mistral-large-2402-v1:0",
           "mistral.mistral-small-2402-v1:0",
@@ -126,6 +127,8 @@ public class AwsBedrockPlugin implements Plugin {
           "mistral.voxtral-mini-3b-2507",
           "mistral.voxtral-small-24b-2507",
           // DeepSeek models
+          "deepseek.v3.2",
+          "deepseek.v3-v1:0",
           "deepseek.r1-v1:0",
           // Google Gemma models
           "google.gemma-3-27b-it",
@@ -149,9 +152,16 @@ public class AwsBedrockPlugin implements Plugin {
           "writer.palmyra-x5-v1:0",
           "writer.palmyra-x4-v1:0",
           // MiniMax models
+          "minimax.minimax-m2.5",
+          "minimax.minimax-m2.1",
           "minimax.minimax-m2",
           // Moonshot models
+          "moonshotai.kimi-k2.5",
           "moonshot.kimi-k2-thinking",
+          // Z.AI models
+          "zai.glm-5",
+          "zai.glm-4.7",
+          "zai.glm-4.7-flash",
           // TwelveLabs models
           "twelvelabs.pegasus-1-2-v1:0");
 
@@ -212,9 +222,24 @@ public String getName() {
       logger.debug("Registered custom AWS Bedrock model: {}", modelId);
     }
 
+    // Register AWS Bedrock embedding models
+    for (String modelId : SUPPORTED_EMBEDDING_MODELS) {
+      AwsBedrockEmbedder embedder = new AwsBedrockEmbedder(modelId, options);
+      actions.add(embedder);
+      logger.debug("Registered AWS Bedrock embedder: {}", modelId);
+    }
+
+    // Register custom embedding models added via customEmbeddingModel()
+    for (String modelId : customEmbeddingModels) {
+      AwsBedrockEmbedder embedder = new AwsBedrockEmbedder(modelId, options);
+      actions.add(embedder);
+      logger.debug("Registered custom AWS Bedrock embedder: {}", modelId);
+    }
+
     logger.info(
-        "AWS Bedrock plugin initialized with {} models in region {} (supports inference profiles)",
+        "AWS Bedrock plugin initialized with {} models and {} embedders in region {} (supports inference profiles)",
         SUPPORTED_MODELS.size() + customModels.size(),
+        SUPPORTED_EMBEDDING_MODELS.size() + customEmbeddingModels.size(),
         options.getRegion());
 
     return actions;
@@ -235,6 +260,19 @@ public AwsBedrockPlugin customModel(String modelId) {
     return this;
   }
 
+  /**
+   * Registers a custom embedding model ID. Use this to work with Bedrock embedding models not in
+   * the default list. Call this method before passing the plugin to Genkit.builder().
+   *
+   * @param modelId the embedding model ID (e.g., "amazon.titan-embed-text-v1")
+   * @return this plugin instance for method chaining
+   */
+  public AwsBedrockPlugin customEmbeddingModel(String modelId) {
+    customEmbeddingModels.add(modelId);
+    logger.debug("Added custom embedding model to be registered: {}", modelId);
+    return this;
+  }
+
   /**
    * Gets the plugin options.
    *
diff --git a/plugins/aws-bedrock/src/main/java/com/google/genkit/plugins/awsbedrock/AwsBedrockSigner.java b/plugins/aws-bedrock/src/main/java/com/google/genkit/plugins/awsbedrock/AwsBedrockSigner.java
new file mode 100644
index 000000000..e6f8b72e8
--- /dev/null
+++ b/plugins/aws-bedrock/src/main/java/com/google/genkit/plugins/awsbedrock/AwsBedrockSigner.java
@@ -0,0 +1,109 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package com.google.genkit.plugins.awsbedrock;
+
+import com.google.genkit.core.GenkitException;
+import okhttp3.MediaType;
+import okhttp3.Request;
+import okhttp3.RequestBody;
+import software.amazon.awssdk.auth.credentials.AwsCredentials;
+import software.amazon.awssdk.auth.signer.Aws4Signer;
+import software.amazon.awssdk.auth.signer.params.Aws4SignerParams;
+import software.amazon.awssdk.http.SdkHttpFullRequest;
+import software.amazon.awssdk.http.SdkHttpMethod;
+
+/**
+ * Helper for AWS SigV4-signing AWS Bedrock runtime HTTP requests. Shared by {@link AwsBedrockModel}
+ * (Converse API) and {@link AwsBedrockEmbedder} (InvokeModel API).
+ */
+final class AwsBedrockSigner {
+
+  private AwsBedrockSigner() {}
+
+  /**
+   * Returns the {@code bedrock-runtime} host for the configured region.
+   *
+   * @param options the plugin options
+   * @return the host name
+   */
+  static String runtimeHost(AwsBedrockPluginOptions options) {
+    return String.format("bedrock-runtime.%s.amazonaws.com", options.getRegion().id());
+  }
+
+  /**
+   * Builds a SigV4-signed OkHttp POST request for the AWS Bedrock runtime.
+   *
+   * @param options the plugin options (region + credentials)
+   * @param host the target host (e.g. from {@link #runtimeHost})
+   * @param path the request path (e.g. {@code /model//invoke})
+   * @param body the JSON request body
+   * @return a signed OkHttp request
+   */
+  static Request signRequest(
+      AwsBedrockPluginOptions options, String host, String path, String body) {
+    try {
+      AwsCredentials credentials = options.getCredentialsProvider().resolveCredentials();
+
+      java.net.URI uri = java.net.URI.create(String.format("https://%s%s", host, path));
+
+      SdkHttpFullRequest httpRequest =
+          SdkHttpFullRequest.builder()
+              .uri(uri)
+              .method(SdkHttpMethod.POST)
+              .putHeader("Content-Type", "application/json")
+              .putHeader("Accept", "application/json")
+              .contentStreamProvider(
+                  () ->
+                      new java.io.ByteArrayInputStream(
+                          body.getBytes(java.nio.charset.StandardCharsets.UTF_8)))
+              .build();
+
+      Aws4Signer signer = Aws4Signer.create();
+      Aws4SignerParams signerParams =
+          Aws4SignerParams.builder()
+              .awsCredentials(credentials)
+              .signingName("bedrock")
+              .signingRegion(options.getRegion())
+              .build();
+
+      SdkHttpFullRequest signedRequest = signer.sign(httpRequest, signerParams);
+
+      // Post the body as raw bytes with no OkHttp media type. The String RequestBody overload
+      // appends "; charset=utf-8" to Content-Type, which the Bedrock InvokeModel API rejects. The
+      // signed Content-Type ("application/json") and Accept headers copied below are the sole
+      // source
+      // of those headers, keeping the request byte-consistent with the SigV4 signature.
+      Request.Builder okHttpRequestBuilder =
+          new Request.Builder()
+              .url(signedRequest.getUri().toURL())
+              .post(
+                  RequestBody.create(
+                      body.getBytes(java.nio.charset.StandardCharsets.UTF_8), (MediaType) null));
+
+      signedRequest
+          .headers()
+          .forEach(
+              (key, values) -> values.forEach(value -> okHttpRequestBuilder.addHeader(key, value)));
+
+      return okHttpRequestBuilder.build();
+    } catch (Exception e) {
+      throw new GenkitException("Failed to sign AWS request", e);
+    }
+  }
+}
diff --git a/plugins/aws-bedrock/src/test/java/com/google/genkit/plugins/awsbedrock/AwsBedrockPluginTest.java b/plugins/aws-bedrock/src/test/java/com/google/genkit/plugins/awsbedrock/AwsBedrockPluginTest.java
index af189c58d..0a1b0bf89 100644
--- a/plugins/aws-bedrock/src/test/java/com/google/genkit/plugins/awsbedrock/AwsBedrockPluginTest.java
+++ b/plugins/aws-bedrock/src/test/java/com/google/genkit/plugins/awsbedrock/AwsBedrockPluginTest.java
@@ -64,6 +64,24 @@ void testSupportedModels() {
         AwsBedrockPlugin.SUPPORTED_MODELS.contains("anthropic.claude-sonnet-4-20250514-v1:0"));
   }
 
+  @Test
+  void testSupportedEmbeddingModels() {
+    assertNotNull(AwsBedrockPlugin.SUPPORTED_EMBEDDING_MODELS);
+    assertTrue(
+        AwsBedrockPlugin.SUPPORTED_EMBEDDING_MODELS.contains("amazon.titan-embed-text-v2:0"));
+  }
+
+  @Test
+  void testRegistersEmbedders() {
+    AwsBedrockPluginOptions options = AwsBedrockPluginOptions.builder().region("us-east-1").build();
+    AwsBedrockPlugin plugin = new AwsBedrockPlugin(options);
+    List> actions = plugin.init();
+    assertTrue(
+        actions.stream()
+            .anyMatch(a -> "aws-bedrock/amazon.titan-embed-text-v2:0".equals(a.getName())),
+        "Should register the Titan embedder");
+  }
+
   @Test
   void testInitializesActions() {
     AwsBedrockPluginOptions options = AwsBedrockPluginOptions.builder().region("us-east-1").build();
diff --git a/plugins/azure-foundry/README.md b/plugins/azure-foundry/README.md
index f6c3c6877..f021cfc01 100644
--- a/plugins/azure-foundry/README.md
+++ b/plugins/azure-foundry/README.md
@@ -105,23 +105,23 @@ Genkit genkit = Genkit.builder()
 ## Supported Models
 
 ### Azure OpenAI Models (Global Standard & Provisioned)
-- **GPT-5**: `gpt-5`, `gpt-5-mini`, `gpt-5-turbo`
-- **o1**: `o1`
-- **o3-mini**: `o3-mini-high`, `o3-mini-medium`, `o3-mini-low`
-- **GPT-4o**: `gpt-4o`, `gpt-4o-mini`
-- **GPT-4**: `gpt-4-turbo`, `gpt-4`, `gpt-35-turbo`
+- **GPT-5.5 / 5.4**: `gpt-5.5`, `gpt-5.4`, `gpt-5.4-mini`, `gpt-5.4-nano`, `gpt-5.4-pro`
+- **GPT-5**: `gpt-5.2`, `gpt-5.1`, `gpt-5`, `gpt-5-mini`, `gpt-5-nano`, `gpt-5-pro`, `gpt-5-codex`
+- **o-series**: `o3`, `o3-pro`, `o3-mini`
+- **GPT-4.1**: `gpt-4.1`, `gpt-4.1-mini`
+- **GPT-4o / 4**: `gpt-4o`, `gpt-4o-mini`, `gpt-4`
 
 ### Azure Models (Sold Directly by Azure)
-- **MAI-DS**: `mai-ds-r1` - Deterministic, precision-focused reasoning
-- **Grok**: `grok-4`, `grok-4-fast-reasoning`, `grok-4-fast-non-reasoning`, `grok-3`, `grok-3-mini`
+- **Grok**: `grok-4`, `grok-4-1-fast-reasoning`, `grok-4-1-fast-non-reasoning`, `grok-code-fast-1`
 - **Llama**: `llama-3-3-70b-instruct`, `llama-4-maverick-17b-128e-instruct-fp8`
-- **DeepSeek**: `deepseek-v3-0324`, `deepseek-v3-1`, `deepseek-r1-0528`
+- **DeepSeek**: `DeepSeek-V3.2`, `DeepSeek-V3.2-Speciale`
+- **Mistral**: `Mistral-Large-3`
 - **GPT-OSS**: `gpt-oss-120b`
 
 ### Partner and Community Models
-- **Claude 4.x**: `claude-opus-4-5`, `claude-opus-4-1`, `claude-sonnet-4-5`, `claude-haiku-4-5`
+- **Claude**: `claude-opus-4-8`, `claude-sonnet-5`, `claude-opus-4-7`, `claude-opus-4-6`, `claude-sonnet-4-6`, `claude-opus-4-5`, `claude-opus-4-1`, `claude-sonnet-4-5`, `claude-haiku-4-5`
 
-> **Note:** Model availability varies by region and Azure subscription. Hub-based projects are limited to gpt-4o, gpt-4o-mini, gpt-4, and gpt-35-turbo. See [Azure AI Foundry Model Region Support](https://learn.microsoft.com/en-us/azure/ai-foundry/agents/concepts/model-region-support) for details.
+> **Note:** Model availability varies by region and Azure subscription. Hub-based projects are limited to gpt-4o, gpt-4o-mini, and gpt-4. See [Azure AI Foundry Model Region Support](https://learn.microsoft.com/en-us/azure/ai-foundry/agents/concepts/model-region-support) for details.
 
 ## Usage Examples
 
diff --git a/plugins/azure-foundry/src/main/java/com/google/genkit/plugins/azurefoundry/AzureFoundryPlugin.java b/plugins/azure-foundry/src/main/java/com/google/genkit/plugins/azurefoundry/AzureFoundryPlugin.java
index 5e3827d5d..ae40cd1e4 100644
--- a/plugins/azure-foundry/src/main/java/com/google/genkit/plugins/azurefoundry/AzureFoundryPlugin.java
+++ b/plugins/azure-foundry/src/main/java/com/google/genkit/plugins/azurefoundry/AzureFoundryPlugin.java
@@ -46,32 +46,43 @@ public class AzureFoundryPlugin implements Plugin {
   public static final List SUPPORTED_MODELS =
       Arrays.asList(
           // Azure OpenAI models (Global Standard & Provisioned)
+          "gpt-5.5",
+          "gpt-5.4",
+          "gpt-5.4-mini",
+          "gpt-5.4-nano",
+          "gpt-5.4-pro",
+          "gpt-5.2",
+          "gpt-5.1",
           "gpt-5",
           "gpt-5-mini",
-          "gpt-5-turbo",
-          "o1",
-          "o3-mini-high",
-          "o3-mini-medium",
-          "o3-mini-low",
+          "gpt-5-nano",
+          "gpt-5-pro",
+          "gpt-5-codex",
+          "o3",
+          "o3-pro",
+          "o3-mini",
+          "gpt-4.1",
+          "gpt-4.1-mini",
           "gpt-4o",
           "gpt-4o-mini",
-          "gpt-4-turbo",
           "gpt-4",
-          "gpt-35-turbo",
           // Azure models sold directly by Azure
-          "mai-ds-r1",
           "grok-4",
-          "grok-4-fast-reasoning",
-          "grok-4-fast-non-reasoning",
-          "grok-3",
-          "grok-3-mini",
+          "grok-4-1-fast-reasoning",
+          "grok-4-1-fast-non-reasoning",
+          "grok-code-fast-1",
           "llama-3-3-70b-instruct",
           "llama-4-maverick-17b-128e-instruct-fp8",
-          "deepseek-v3-0324",
-          "deepseek-v3-1",
-          "deepseek-r1-0528",
+          "DeepSeek-V3.2",
+          "DeepSeek-V3.2-Speciale",
+          "Mistral-Large-3",
           "gpt-oss-120b",
-          // Partner and community models
+          // Partner and community models (Anthropic Claude)
+          "claude-opus-4-8",
+          "claude-sonnet-5",
+          "claude-opus-4-7",
+          "claude-opus-4-6",
+          "claude-sonnet-4-6",
           "claude-opus-4-5",
           "claude-opus-4-1",
           "claude-sonnet-4-5",
diff --git a/plugins/cohere/README.md b/plugins/cohere/README.md
index 890573721..a0b4a2667 100644
--- a/plugins/cohere/README.md
+++ b/plugins/cohere/README.md
@@ -18,6 +18,16 @@ This plugin provides integration with Cohere models.
 - `command-r-08-2024` - Balanced model for complex workflows (128K context)
 - `command-r-plus-08-2024` - Enhanced model for complex RAG and multi-step tool use (128K context)
 
+## Embeddings
+
+Cohere embedding models are available via the OpenAI-compatible endpoint:
+
+- `embed-v4.0` - Latest multimodal embeddings (text + images)
+- `embed-multilingual-v3.0`
+- `embed-english-v3.0`
+
+Register additional embedding models with `customEmbeddingModel(...)`.
+
 ## Using Custom Models
 
 If you need to use a model not in the default list (e.g., a newer model release), register it using `customModel()`:
diff --git a/plugins/cohere/src/main/java/com/google/genkit/plugins/cohere/CoherePlugin.java b/plugins/cohere/src/main/java/com/google/genkit/plugins/cohere/CoherePlugin.java
index 2ed5f0796..1845994e0 100644
--- a/plugins/cohere/src/main/java/com/google/genkit/plugins/cohere/CoherePlugin.java
+++ b/plugins/cohere/src/main/java/com/google/genkit/plugins/cohere/CoherePlugin.java
@@ -20,6 +20,7 @@
 
 import com.google.genkit.core.Action;
 import com.google.genkit.core.Plugin;
+import com.google.genkit.plugins.compatoai.CompatOAIEmbedder;
 import com.google.genkit.plugins.compatoai.CompatOAIModel;
 import com.google.genkit.plugins.compatoai.CompatOAIPluginOptions;
 import java.util.ArrayList;
@@ -40,13 +41,23 @@ public class CoherePlugin implements Plugin {
   /** Supported Cohere models. */
   public static final List SUPPORTED_MODELS =
       Arrays.asList(
+          // Command A family
+          "command-a-plus-05-2026",
+          "command-a-reasoning-08-2025",
+          "command-a-vision-07-2025",
           "command-a-03-2025",
+          // Command R family
           "command-r7b-12-2024",
           "command-r-08-2024",
           "command-r-plus-08-2024");
 
+  /** Supported Cohere embedding models (reachable via the OpenAI-compatible endpoint). */
+  public static final List SUPPORTED_EMBEDDING_MODELS =
+      Arrays.asList("embed-v4.0", "embed-multilingual-v3.0", "embed-english-v3.0");
+
   private final CompatOAIPluginOptions options;
   private final List customModels = new ArrayList<>();
+  private final List customEmbeddingModels = new ArrayList<>();
 
   /** Creates a CoherePlugin with default options (using COHERE_API_KEY environment variable). */
   public CoherePlugin() {
@@ -128,8 +139,26 @@ public String getName() {
       logger.debug("Created custom Cohere model: {}", modelName);
     }
 
+    // Register Cohere embedding models
+    for (String modelName : SUPPORTED_EMBEDDING_MODELS) {
+      CompatOAIEmbedder embedder =
+          new CompatOAIEmbedder("cohere/" + modelName, modelName, "Cohere " + modelName, options);
+      actions.add(embedder);
+      logger.debug("Created Cohere embedder: {}", modelName);
+    }
+
+    // Register custom embedding models added via customEmbeddingModel()
+    for (String modelName : customEmbeddingModels) {
+      CompatOAIEmbedder embedder =
+          new CompatOAIEmbedder("cohere/" + modelName, modelName, "Cohere " + modelName, options);
+      actions.add(embedder);
+      logger.debug("Created custom Cohere embedder: {}", modelName);
+    }
+
     logger.info(
-        "Cohere plugin initialized with {} models", SUPPORTED_MODELS.size() + customModels.size());
+        "Cohere plugin initialized with {} models and {} embedders",
+        SUPPORTED_MODELS.size() + customModels.size(),
+        SUPPORTED_EMBEDDING_MODELS.size() + customEmbeddingModels.size());
 
     return actions;
   }
@@ -147,6 +176,19 @@ public CoherePlugin customModel(String modelName) {
     return this;
   }
 
+  /**
+   * Registers a custom embedding model name. Use this to work with embedding models not in the
+   * default list. Call this method before passing the plugin to Genkit.builder().
+   *
+   * @param modelName the embedding model name (e.g., "embed-english-light-v3.0")
+   * @return this plugin instance for method chaining
+   */
+  public CoherePlugin customEmbeddingModel(String modelName) {
+    customEmbeddingModels.add(modelName);
+    logger.debug("Added custom embedding model to be registered: {}", modelName);
+    return this;
+  }
+
   /**
    * Gets the plugin options.
    *
diff --git a/plugins/cohere/src/test/java/com/google/genkit/plugins/cohere/CoherePluginTest.java b/plugins/cohere/src/test/java/com/google/genkit/plugins/cohere/CoherePluginTest.java
index e2b611192..0a9a80d02 100644
--- a/plugins/cohere/src/test/java/com/google/genkit/plugins/cohere/CoherePluginTest.java
+++ b/plugins/cohere/src/test/java/com/google/genkit/plugins/cohere/CoherePluginTest.java
@@ -92,6 +92,26 @@ void testSupportedModels() {
     assertTrue(CoherePlugin.SUPPORTED_MODELS.contains("command-r-plus-08-2024"));
   }
 
+  @Test
+  void testSupportedEmbeddingModels() {
+    assertNotNull(CoherePlugin.SUPPORTED_EMBEDDING_MODELS);
+    assertTrue(CoherePlugin.SUPPORTED_EMBEDDING_MODELS.contains("embed-v4.0"));
+  }
+
+  @Test
+  void testRegistersEmbedders() {
+    CoherePlugin plugin =
+        new CoherePlugin(
+            CompatOAIPluginOptions.builder()
+                .apiKey("test-key")
+                .baseUrl("https://api.cohere.ai/compatibility/v1")
+                .build());
+    List> actions = plugin.init();
+    assertTrue(
+        actions.stream().anyMatch(a -> "cohere/embed-v4.0".equals(a.getName())),
+        "Should register the embed-v4.0 embedder");
+  }
+
   @Test
   void testCustomModel() {
     CoherePlugin plugin =
diff --git a/plugins/compat-oai/src/main/java/com/google/genkit/plugins/compatoai/CompatOAIEmbedder.java b/plugins/compat-oai/src/main/java/com/google/genkit/plugins/compatoai/CompatOAIEmbedder.java
new file mode 100644
index 000000000..5c2336f4a
--- /dev/null
+++ b/plugins/compat-oai/src/main/java/com/google/genkit/plugins/compatoai/CompatOAIEmbedder.java
@@ -0,0 +1,178 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package com.google.genkit.plugins.compatoai;
+
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.fasterxml.jackson.databind.node.ArrayNode;
+import com.fasterxml.jackson.databind.node.ObjectNode;
+import com.google.genkit.ai.Document;
+import com.google.genkit.ai.EmbedRequest;
+import com.google.genkit.ai.EmbedResponse;
+import com.google.genkit.ai.Embedder;
+import com.google.genkit.ai.EmbedderInfo;
+import com.google.genkit.core.ActionContext;
+import com.google.genkit.core.GenkitException;
+import java.io.IOException;
+import java.util.ArrayList;
+import java.util.List;
+import java.util.concurrent.TimeUnit;
+import okhttp3.*;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+/**
+ * Embedder implementation for any provider that exposes an OpenAI-compatible {@code /embeddings}
+ * endpoint (e.g. Cohere, Mistral).
+ *
+ * 
Sends a {@code POST {baseUrl}/embeddings} request with a JSON body of the form {@code
+ * {"model": "...", "input": ["text", ...]}} and parses the {@code data[].embedding} arrays.
+ */
+public class CompatOAIEmbedder extends Embedder {
+
+  private static final Logger logger = LoggerFactory.getLogger(CompatOAIEmbedder.class);
+  private static final MediaType JSON_MEDIA_TYPE = MediaType.parse("application/json");
+
+  private final String apiModelName;
+  private final CompatOAIPluginOptions options;
+  private final OkHttpClient client;
+  private final ObjectMapper objectMapper;
+
+  /**
+   * Creates a new CompatOAIEmbedder.
+   *
+   * @param modelName the Genkit embedder name (e.g. "cohere/embed-v4.0")
+   * @param apiModelName the model name sent to the API (e.g. "embed-v4.0")
+   * @param label the display label (e.g. "Cohere embed-v4.0")
+   * @param options the plugin options
+   */
+  public CompatOAIEmbedder(
+      String modelName, String apiModelName, String label, CompatOAIPluginOptions options) {
+    super(
+        modelName,
+        createEmbedderInfo(label),
+        (ctx, req) -> {
+          throw new GenkitException("Handler not initialized");
+        });
+    this.apiModelName = apiModelName;
+    this.options = options;
+    this.objectMapper = new ObjectMapper();
+    this.client =
+        new OkHttpClient.Builder()
+            .connectTimeout(options.getTimeout(), TimeUnit.SECONDS)
+            .readTimeout(options.getTimeout(), TimeUnit.SECONDS)
+            .writeTimeout(options.getTimeout(), TimeUnit.SECONDS)
+            .build();
+  }
+
+  private static EmbedderInfo createEmbedderInfo(String label) {
+    EmbedderInfo info = new EmbedderInfo();
+    info.setLabel(label);
+    return info;
+  }
+
+  @Override
+  public EmbedResponse run(ActionContext context, EmbedRequest request) {
+    if (request == null) {
+      throw new GenkitException(
+          "Embed request is required. Please provide an input with documents to embed.");
+    }
+    if (request.getDocuments() == null || request.getDocuments().isEmpty()) {
+      throw new GenkitException("Embed request must contain at least one document to embed.");
+    }
+    try {
+      return callApi(request);
+    } catch (IOException e) {
+      throw new GenkitException("Embedding API call failed", e);
+    }
+  }
+
+  private String buildUrl() {
+    StringBuilder url = new StringBuilder(options.getBaseUrl());
+    url.append("/embeddings");
+    if (options.getQueryParams() != null && !options.getQueryParams().isEmpty()) {
+      url.append("?");
+      boolean first = true;
+      for (java.util.Map.Entry entry : options.getQueryParams().entrySet()) {
+        if (!first) {
+          url.append("&");
+        }
+        url.append(entry.getKey()).append("=").append(entry.getValue());
+        first = false;
+      }
+    }
+    return url.toString();
+  }
+
+  private EmbedResponse callApi(EmbedRequest request) throws IOException {
+    ObjectNode requestBody = objectMapper.createObjectNode();
+    requestBody.put("model", apiModelName);
+
+    ArrayNode input = requestBody.putArray("input");
+    for (Document doc : request.getDocuments()) {
+      String text = doc.text();
+      if (text == null || text.isEmpty()) {
+        logger.warn("Document has empty text, skipping");
+        continue;
+      }
+      input.add(text);
+    }
+    if (input.isEmpty()) {
+      throw new GenkitException("No valid documents to embed - all documents had empty text");
+    }
+
+    Request.Builder requestBuilder =
+        new Request.Builder()
+            .url(buildUrl())
+            .header("Authorization", "Bearer " + options.getApiKey())
+            .header("Content-Type", "application/json")
+            .post(RequestBody.create(requestBody.toString(), JSON_MEDIA_TYPE));
+    if (options.getOrganization() != null) {
+      requestBuilder.header("OpenAI-Organization", options.getOrganization());
+    }
+
+    try (Response response = client.newCall(requestBuilder.build()).execute()) {
+      if (!response.isSuccessful()) {
+        String errorBody = response.body() != null ? response.body().string() : "No error body";
+        throw new GenkitException("Embedding API error: " + response.code() + " - " + errorBody);
+      }
+      return parseResponse(response.body().string());
+    }
+  }
+
+  private EmbedResponse parseResponse(String responseBody) throws IOException {
+    JsonNode root = objectMapper.readTree(responseBody);
+    List embeddings = new ArrayList<>();
+
+    JsonNode dataNode = root.get("data");
+    if (dataNode != null && dataNode.isArray()) {
+      for (JsonNode item : dataNode) {
+        JsonNode embeddingNode = item.get("embedding");
+        if (embeddingNode != null && embeddingNode.isArray()) {
+          float[] values = new float[embeddingNode.size()];
+          for (int i = 0; i < embeddingNode.size(); i++) {
+            values[i] = (float) embeddingNode.get(i).asDouble();
+          }
+          embeddings.add(new EmbedResponse.Embedding(values));
+        }
+      }
+    }
+    return new EmbedResponse(embeddings);
+  }
+}
diff --git a/plugins/deepseek/src/main/java/com/google/genkit/plugins/deepseek/DeepSeekPlugin.java b/plugins/deepseek/src/main/java/com/google/genkit/plugins/deepseek/DeepSeekPlugin.java
index 647cce80f..f61e6ec81 100644
--- a/plugins/deepseek/src/main/java/com/google/genkit/plugins/deepseek/DeepSeekPlugin.java
+++ b/plugins/deepseek/src/main/java/com/google/genkit/plugins/deepseek/DeepSeekPlugin.java
@@ -39,7 +39,13 @@ public class DeepSeekPlugin implements Plugin {
 
   /** Supported DeepSeek models. */
   public static final List SUPPORTED_MODELS =
-      Arrays.asList("deepseek-chat", "deepseek-reasoner");
+      Arrays.asList(
+          // DeepSeek V4 (current)
+          "deepseek-v4-pro",
+          "deepseek-v4-flash",
+          // Legacy aliases (route to V4; deprecated, scheduled for retirement 2026-07-24)
+          "deepseek-chat",
+          "deepseek-reasoner");
 
   private final CompatOAIPluginOptions options;
   private final List customModels = new ArrayList<>();
diff --git a/plugins/google-genai/src/main/java/com/google/genkit/plugins/googlegenai/GeminiEmbedder.java b/plugins/google-genai/src/main/java/com/google/genkit/plugins/googlegenai/GeminiEmbedder.java
index c097eaa32..42c940ef4 100644
--- a/plugins/google-genai/src/main/java/com/google/genkit/plugins/googlegenai/GeminiEmbedder.java
+++ b/plugins/google-genai/src/main/java/com/google/genkit/plugins/googlegenai/GeminiEmbedder.java
@@ -46,7 +46,7 @@ public class GeminiEmbedder extends Embedder {
   /**
    * Creates a new GeminiEmbedder.
    *
-   * @param modelName the embedding model name (e.g., "text-embedding-004", "gemini-embedding-001")
+   * @param modelName the embedding model name (e.g., "gemini-embedding-2", "gemini-embedding-001")
    * @param options the plugin options
    */
   public GeminiEmbedder(String modelName, GoogleGenAIPluginOptions options) {
diff --git a/plugins/google-genai/src/main/java/com/google/genkit/plugins/googlegenai/GoogleGenAIPlugin.java b/plugins/google-genai/src/main/java/com/google/genkit/plugins/googlegenai/GoogleGenAIPlugin.java
index e53c87941..a4b1fd35b 100644
--- a/plugins/google-genai/src/main/java/com/google/genkit/plugins/googlegenai/GoogleGenAIPlugin.java
+++ b/plugins/google-genai/src/main/java/com/google/genkit/plugins/googlegenai/GoogleGenAIPlugin.java
@@ -32,9 +32,9 @@
  * 
This plugin provides access to Google's Gemini models for:
  *
  * 
    
- *   
  • Text generation (Gemini 2.0, 2.5, 3.0 series)
+ *   
  • Text generation (Gemini 3.5, 3.1, 2.5 series)
  *   
  • Multimodal content (images, video, audio)
- *   
  • Embeddings (text-embedding-004, gemini-embedding-001)
+ *   
  • Embeddings (gemini-embedding-2, gemini-embedding-001)
  *   
  • Function calling/tools
  * 

  *
@@ -67,7 +67,7 @@
  * // Generate content
  * GenerateResponse response = genkit.generate(
  *     GenerateOptions.builder()
- *         .model("googleai/gemini-2.0-flash")
+ *         .model("googleai/gemini-2.5-flash")
  *         .prompt("Hello, world!")
  *         .build());
  * }

@@ -79,33 +79,26 @@ public class GoogleGenAIPlugin implements Plugin {
   /** Supported Gemini models for text/multimodal generation. */
   public static final List SUPPORTED_MODELS =
       Arrays.asList(
-          // Gemini 3.0 series
-          "gemini-3-pro-preview",
+          // Gemini 3.5 / 3.1 series
+          "gemini-3.5-flash",
+          "gemini-3.1-pro-preview",
+          "gemini-3.1-flash-lite",
           "gemini-3-flash-preview",
           // Gemini 2.5 series
           "gemini-2.5-pro",
           "gemini-2.5-flash",
           "gemini-2.5-flash-lite",
-          // Gemini 2.0 series
-          "gemini-2.0-flash",
-          "gemini-2.0-flash-lite",
-          // Gemini 1.5 series (still widely used)
-          "gemini-1.5-pro",
-          "gemini-1.5-flash",
-          "gemini-1.5-flash-8b",
           // Gemma models
-          "gemma-3-12b-it",
-          "gemma-3-27b-it",
-          "gemma-3-4b-it",
-          "gemma-3-1b-it",
-          "gemma-3n-e4b-it");
+          "gemma-4-31b-it",
+          "gemma-4-26b-a4b-it");
 
   /** Supported embedding models. */
   public static final List SUPPORTED_EMBEDDING_MODELS =
       Arrays.asList(
-          "text-embedding-004",
-          "text-embedding-005",
+          "gemini-embedding-2",
           "gemini-embedding-001",
+          // Vertex AI embedding models
+          "text-embedding-005",
           "text-multilingual-embedding-002");
 
   /**
@@ -113,20 +106,28 @@ public class GoogleGenAIPlugin implements Plugin {
    * Gemini Developer API. imagen-3.0-* models require Vertex AI.
    */
   public static final List SUPPORTED_IMAGE_MODELS =
-      Arrays.asList("imagen-4.0-generate-001", "imagen-4.0-fast-generate-001");
+      Arrays.asList(
+          "imagen-4.0-generate-001",
+          "imagen-4.0-fast-generate-001",
+          "imagen-4.0-ultra-generate-001");
 
   /** Supported TTS models. */
   public static final List SUPPORTED_TTS_MODELS =
-      Arrays.asList("gemini-2.5-flash-preview-tts", "gemini-2.5-pro-preview-tts");
+      Arrays.asList(
+          "gemini-3.1-flash-tts-preview",
+          "gemini-2.5-flash-preview-tts",
+          "gemini-2.5-pro-preview-tts");
 
   /** Supported video generation models (Veo). */
   public static final List SUPPORTED_VEO_MODELS =
       Arrays.asList(
-          "veo-2.0-generate-001",
-          "veo-3.0-generate-001",
-          "veo-3.0-fast-generate-001",
           "veo-3.1-generate-preview",
-          "veo-3.1-fast-generate-preview");
+          "veo-3.1-fast-generate-preview",
+          "veo-3.1-lite-generate-preview");
+
+  /** Supported Gemini Omni video generation/editing models (Interactions API). */
+  public static final List SUPPORTED_OMNI_MODELS =
+      Arrays.asList("gemini-omni-flash-preview");
 
   private final GoogleGenAIPluginOptions options;
   private final List customModels = new ArrayList<>();
@@ -134,6 +135,7 @@ public class GoogleGenAIPlugin implements Plugin {
   private final List customImageModels = new ArrayList<>();
   private final List customTtsModels = new ArrayList<>();
   private final List customVeoModels = new ArrayList<>();
+  private final List customOmniModels = new ArrayList<>();
 
   /**
    * Creates a GoogleGenAIPlugin with default options. Reads API key from GOOGLE_API_KEY or
@@ -276,14 +278,30 @@ public String getName() {
       logger.debug("Created custom Veo model: {}", modelName);
     }
 
+    // Register Gemini Omni video generation/editing models (Interactions API)
+    for (String modelName : SUPPORTED_OMNI_MODELS) {
+      OmniModel model = new OmniModel(modelName, options);
+      actions.add(model);
+      logger.debug("Created Omni model: {}", modelName);
+    }
+
+    // Register custom Omni models
+    for (String modelName : customOmniModels) {
+      OmniModel model = new OmniModel(modelName, options);
+      actions.add(model);
+      logger.debug("Created custom Omni model: {}", modelName);
+    }
+
     String backend = options.isVertexAI() ? "Vertex AI" : "Gemini Developer API";
     logger.info(
-        "Google GenAI plugin initialized with {} models, {} embedders, {} image models, {} TTS models, and {} video models using {}",
+        "Google GenAI plugin initialized with {} models, {} embedders, {} image models, {} TTS"
+            + " models, {} video models, and {} omni models using {}",
         SUPPORTED_MODELS.size() + customModels.size(),
         SUPPORTED_EMBEDDING_MODELS.size() + customEmbeddingModels.size(),
         SUPPORTED_IMAGE_MODELS.size() + customImageModels.size(),
         SUPPORTED_TTS_MODELS.size() + customTtsModels.size(),
         SUPPORTED_VEO_MODELS.size() + customVeoModels.size(),
+        SUPPORTED_OMNI_MODELS.size() + customOmniModels.size(),
         backend);
 
     return actions;
@@ -354,6 +372,19 @@ public GoogleGenAIPlugin customVeoModel(String modelName) {
     return this;
   }
 
+  /**
+   * Registers a custom Gemini Omni model name. Use this to work with Omni (Interactions API) models
+   * not in the default list. Call this method before passing the plugin to Genkit.builder().
+   *
+   * @param modelName the Omni model name (e.g., "gemini-omni-pro-preview")
+   * @return this plugin instance for method chaining
+   */
+  public GoogleGenAIPlugin customOmniModel(String modelName) {
+    customOmniModels.add(modelName);
+    logger.debug("Added custom Omni model to be registered: {}", modelName);
+    return this;
+  }
+
   /**
    * Gets the plugin options.
    *
diff --git a/plugins/google-genai/src/main/java/com/google/genkit/plugins/googlegenai/ImagenModel.java b/plugins/google-genai/src/main/java/com/google/genkit/plugins/googlegenai/ImagenModel.java
index 5c647665f..1dc1c9cfc 100644
--- a/plugins/google-genai/src/main/java/com/google/genkit/plugins/googlegenai/ImagenModel.java
+++ b/plugins/google-genai/src/main/java/com/google/genkit/plugins/googlegenai/ImagenModel.java
@@ -73,7 +73,7 @@ public class ImagenModel implements Model {
   /**
    * Creates a new ImagenModel.
    *
-   * @param modelName the model name (e.g., "imagen-3.0-generate-002")
+   * @param modelName the model name (e.g., "imagen-4.0-generate-001")
    * @param options the plugin options
    */
   public ImagenModel(String modelName, GoogleGenAIPluginOptions options) {
diff --git a/plugins/google-genai/src/main/java/com/google/genkit/plugins/googlegenai/OmniModel.java b/plugins/google-genai/src/main/java/com/google/genkit/plugins/googlegenai/OmniModel.java
new file mode 100644
index 000000000..0352ce4ce
--- /dev/null
+++ b/plugins/google-genai/src/main/java/com/google/genkit/plugins/googlegenai/OmniModel.java
@@ -0,0 +1,397 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package com.google.genkit.plugins.googlegenai;
+
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.fasterxml.jackson.databind.node.ArrayNode;
+import com.fasterxml.jackson.databind.node.ObjectNode;
+import com.google.genkit.ai.Candidate;
+import com.google.genkit.ai.FinishReason;
+import com.google.genkit.ai.Media;
+import com.google.genkit.ai.Message;
+import com.google.genkit.ai.Model;
+import com.google.genkit.ai.ModelInfo;
+import com.google.genkit.ai.ModelRequest;
+import com.google.genkit.ai.ModelResponse;
+import com.google.genkit.ai.ModelResponseChunk;
+import com.google.genkit.ai.Part;
+import com.google.genkit.ai.Role;
+import com.google.genkit.core.ActionContext;
+import com.google.genkit.core.GenkitException;
+import java.net.URI;
+import java.net.http.HttpClient;
+import java.net.http.HttpRequest;
+import java.net.http.HttpResponse;
+import java.time.Duration;
+import java.util.ArrayList;
+import java.util.HashMap;
+import java.util.List;
+import java.util.Map;
+import java.util.Set;
+import java.util.function.Consumer;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+/**
+ * Gemini Omni video generation and editing model using the Gemini Interactions API.
+ *
+ * 
Gemini Omni ({@code gemini-omni-flash-preview}) natively processes text, image, audio, and
+ * video and produces video (with audio). It is served by the Interactions API ({@code POST
+ * /v1beta/interactions}) rather than {@code generateContent}, and supports conversational
+ * editing: each turn can build on a previous result.
+ *
+ * 
Conversational editing. The Interactions API is stateful server-side. This model
+ * bridges it to Genkit's stateless {@link Model} interface as follows:
+ *
+ * 
    
+ *   
  • On the first turn, send a prompt (and optional media). The response's {@code custom} map
+ *       carries the returned {@code interactionId}.
+ *   
  • To iteratively edit, pass that id back as the {@code previousInteractionId} config option
+ *       on the next {@code generate} call, with the edit instruction as the prompt. The model
+ *       applies the change while preserving elements you did not mention.
+ * 

+ *
+ * 
Supported config options (via the request config map): {@code previousInteractionId}, {@code
+ * aspectRatio} (e.g. "16:9"), {@code duration} (e.g. "10s"), {@code delivery} ("inline" (default) |
+ * "uri"), {@code task} ("text_to_video" | "image_to_video"), {@code thinkingLevel}, {@code
+ * maxOutputTokens}.
+ *
+ * 
Only the Gemini Developer API (API key) is supported; Vertex AI is not.
+ *
+ * 
Note: the Interactions API is in preview; request/response field shapes may evolve.
+ */
+public class OmniModel implements Model {
+
+  private static final Logger logger = LoggerFactory.getLogger(OmniModel.class);
+
+  private final String modelName;
+  private final GoogleGenAIPluginOptions options;
+  private final HttpClient httpClient;
+  private final ObjectMapper objectMapper;
+  private final ModelInfo info;
+
+  /**
+   * Creates a new OmniModel.
+   *
+   * @param modelName the model name (e.g., "gemini-omni-flash-preview")
+   * @param options the plugin options
+   */
+  public OmniModel(String modelName, GoogleGenAIPluginOptions options) {
+    this.modelName = modelName;
+    this.options = options;
+    this.objectMapper = new ObjectMapper();
+    this.httpClient =
+        HttpClient.newBuilder().connectTimeout(Duration.ofSeconds(options.getTimeout())).build();
+    this.info = createModelInfo();
+  }
+
+  private ModelInfo createModelInfo() {
+    ModelInfo info = new ModelInfo();
+    info.setLabel("Google AI " + modelName);
+
+    ModelInfo.ModelCapabilities caps = new ModelInfo.ModelCapabilities();
+    caps.setMultiturn(true); // conversational editing
+    caps.setMedia(true); // accepts text/image/audio/video input
+    caps.setTools(false);
+    caps.setSystemRole(false);
+    caps.setOutput(Set.of("media"));
+    info.setSupports(caps);
+
+    return info;
+  }
+
+  @Override
+  public String getName() {
+    return "googleai/" + modelName;
+  }
+
+  @Override
+  public ModelInfo getInfo() {
+    return info;
+  }
+
+  @Override
+  public boolean supportsStreaming() {
+    return false;
+  }
+
+  @Override
+  public ModelResponse run(ActionContext context, ModelRequest request) {
+    if (options.isVertexAI()) {
+      throw new GenkitException(
+          "Gemini Omni (Interactions API) is not supported on Vertex AI. Use the Gemini Developer"
+              + " API with an API key.");
+    }
+    try {
+      return callInteractions(request);
+    } catch (GenkitException e) {
+      throw e;
+    } catch (Exception e) {
+      throw new GenkitException("Gemini Omni API call failed: " + e.getMessage(), e);
+    }
+  }
+
+  @Override
+  public ModelResponse run(
+      ActionContext context, ModelRequest request, Consumer streamCallback) {
+    // The Interactions API is not streamed; return the full result.
+    return run(context, request);
+  }
+
+  private ModelResponse callInteractions(ModelRequest request) throws Exception {
+    Map config = request.getConfig();
+    String previousInteractionId = configString(config, "previousInteractionId", null);
+    String delivery = configString(config, "delivery", null); // "inline" (default) | "uri"
+    String aspectRatio = configString(config, "aspectRatio", null); // e.g. "16:9"
+    String duration = configString(config, "duration", null); // e.g. "10s"
+    String thinkingLevel = configString(config, "thinkingLevel", "high");
+    int maxOutputTokens = configInt(config, "maxOutputTokens", 65536);
+    boolean isEdit = previousInteractionId != null && !previousInteractionId.isEmpty();
+
+    // Collect the latest user prompt text and any media parts.
+    String promptText = latestUserText(request);
+    List mediaParts = latestUserMedia(request);
+    String task =
+        configString(config, "task", mediaParts.isEmpty() ? "text_to_video" : "image_to_video");
+
+    ObjectNode body = objectMapper.createObjectNode();
+    body.put("model", "models/" + modelName);
+
+    if (isEdit) {
+      // Conversational editing turn: reference the prior interaction.
+      body.put("previous_interaction_id", previousInteractionId);
+      body.put("input", promptText);
+    } else if (mediaParts.isEmpty()) {
+      // Text-to-video: input is a plain prompt string.
+      body.put("input", promptText);
+    } else {
+      // Media-conditioned generation: input is an array of typed objects.
+      ArrayNode input = body.putArray("input");
+      if (promptText != null && !promptText.isEmpty()) {
+        ObjectNode textNode = input.addObject();
+        textNode.put("type", "text");
+        textNode.put("text", promptText);
+      }
+      for (Part part : mediaParts) {
+        input.add(mediaToInput(part.getMedia()));
+      }
+    }
+
+    // Every turn requests video output.
+    body.putArray("response_modalities").add("video");
+    ObjectNode responseFormat = body.putObject("response_format");
+    responseFormat.put("type", "video");
+    if (aspectRatio != null) {
+      responseFormat.put("aspect_ratio", aspectRatio);
+    }
+    if (duration != null) {
+      responseFormat.put("duration", duration);
+    }
+    if (delivery != null) {
+      // Supported values: "inline" (base64, default) or "uri".
+      responseFormat.put("delivery", delivery);
+    }
+
+    // Generation config applies to the initial generation turn.
+    if (!isEdit) {
+      ObjectNode generationConfig = body.putObject("generation_config");
+      generationConfig.put("max_output_tokens", maxOutputTokens);
+      if (thinkingLevel != null && !thinkingLevel.isEmpty()) {
+        generationConfig.put("thinking_level", thinkingLevel);
+      }
+      ObjectNode videoConfig = generationConfig.putObject("video_config");
+      videoConfig.put("task", task);
+    }
+
+    HttpRequest httpRequest =
+        HttpRequest.newBuilder()
+            .uri(URI.create("https://generativelanguage.googleapis.com/v1beta/interactions"))
+            .timeout(Duration.ofSeconds(options.getTimeout()))
+            .header("x-goog-api-key", options.getApiKey())
+            .header("Content-Type", "application/json")
+            .POST(HttpRequest.BodyPublishers.ofString(body.toString()))
+            .build();
+
+    HttpResponse response =
+        httpClient.send(httpRequest, HttpResponse.BodyHandlers.ofString());
+    if (response.statusCode() < 200 || response.statusCode() >= 300) {
+      throw new GenkitException(
+          "Gemini Omni API error: " + response.statusCode() + " - " + response.body());
+    }
+
+    return parseResponse(objectMapper.readTree(response.body()));
+  }
+
+  private ObjectNode mediaToInput(Media media) {
+    // Best-effort mapping of inline media to an Interactions API input object.
+    ObjectNode node = objectMapper.createObjectNode();
+    String contentType = media.getContentType() != null ? media.getContentType() : "";
+    String type = contentType.startsWith("video/") ? "video" : "image";
+    node.put("type", type);
+    String url = media.getUrl();
+    ObjectNode source = node.putObject("source");
+    if (url != null && url.startsWith("data:")) {
+      String[] parts = url.split(",", 2);
+      String mediaType = parts[0].substring(5, parts[0].indexOf(';'));
+      source.put("type", "base64");
+      source.put("media_type", mediaType);
+      source.put("data", parts.length > 1 ? parts[1] : "");
+    } else {
+      source.put("type", "uri");
+      source.put("uri", url);
+    }
+    return node;
+  }
+
+  private ModelResponse parseResponse(JsonNode root) {
+    ModelResponse modelResponse = new ModelResponse();
+    List candidates = new ArrayList<>();
+    Candidate candidate = new Candidate();
+
+    Message message = new Message();
+    message.setRole(Role.MODEL);
+    List parts = new ArrayList<>();
+
+    // Walk the interaction steps and collect video output from the model_output step(s).
+    JsonNode steps = root.get("steps");
+    if (steps != null && steps.isArray()) {
+      for (JsonNode step : steps) {
+        if (!"model_output".equals(step.path("type").asText())) {
+          continue;
+        }
+        JsonNode content = step.get("content");
+        if (content == null || !content.isArray()) {
+          continue;
+        }
+        for (JsonNode item : content) {
+          if (!"video".equals(item.path("type").asText())) {
+            continue;
+          }
+          String mimeType = item.path("mime_type").asText("video/mp4");
+          String url;
+          if (item.hasNonNull("data")) {
+            url = "data:" + mimeType + ";base64," + item.get("data").asText();
+          } else if (item.hasNonNull("uri")) {
+            url = item.get("uri").asText();
+          } else {
+            continue;
+          }
+          Part videoPart = new Part();
+          videoPart.setMedia(new Media(mimeType, url));
+          parts.add(videoPart);
+        }
+      }
+    }
+
+    message.setContent(parts);
+    candidate.setMessage(message);
+    candidate.setFinishReason(FinishReason.STOP);
+
+    // Expose the interaction id so the caller can continue editing (previousInteractionId).
+    String interactionId = root.path("id").asText(null);
+    String status = root.path("status").asText(null);
+    if (interactionId != null) {
+      Map custom = new HashMap<>();
+      custom.put("interactionId", interactionId);
+      if (status != null) {
+        custom.put("status", status);
+      }
+      candidate.setCustom(custom);
+      modelResponse.setCustom(custom);
+    }
+
+    if (parts.isEmpty()) {
+      logger.warn("Gemini Omni response contained no video output (status: {})", status);
+    }
+
+    candidates.add(candidate);
+    modelResponse.setCandidates(candidates);
+    return modelResponse;
+  }
+
+  private static String configString(Map config, String key, String defaultValue) {
+    if (config == null) {
+      return defaultValue;
+    }
+    Object value = config.get(key);
+    return value != null ? value.toString() : defaultValue;
+  }
+
+  private static int configInt(Map config, String key, int defaultValue) {
+    if (config == null) {
+      return defaultValue;
+    }
+    Object value = config.get(key);
+    if (value instanceof Number) {
+      return ((Number) value).intValue();
+    }
+    if (value != null) {
+      try {
+        return Integer.parseInt(value.toString());
+      } catch (NumberFormatException e) {
+        return defaultValue;
+      }
+    }
+    return defaultValue;
+  }
+
+  private String latestUserText(ModelRequest request) {
+    if (request.getMessages() == null) {
+      return "";
+    }
+    for (int i = request.getMessages().size() - 1; i >= 0; i--) {
+      Message message = request.getMessages().get(i);
+      if (message.getRole() != Role.USER) {
+        continue;
+      }
+      StringBuilder sb = new StringBuilder();
+      for (Part part : message.getContent()) {
+        if (part.getText() != null) {
+          if (sb.length() > 0) {
+            sb.append("\n");
+          }
+          sb.append(part.getText());
+        }
+      }
+      return sb.toString();
+    }
+    return "";
+  }
+
+  private List latestUserMedia(ModelRequest request) {
+    List media = new ArrayList<>();
+    if (request.getMessages() == null) {
+      return media;
+    }
+    for (int i = request.getMessages().size() - 1; i >= 0; i--) {
+      Message message = request.getMessages().get(i);
+      if (message.getRole() != Role.USER) {
+        continue;
+      }
+      for (Part part : message.getContent()) {
+        if (part.getMedia() != null) {
+          media.add(part);
+        }
+      }
+      return media;
+    }
+    return media;
+  }
+}
diff --git a/plugins/google-genai/src/main/java/com/google/genkit/plugins/googlegenai/TtsModel.java b/plugins/google-genai/src/main/java/com/google/genkit/plugins/googlegenai/TtsModel.java
index 34642b08a..8c046f544 100644
--- a/plugins/google-genai/src/main/java/com/google/genkit/plugins/googlegenai/TtsModel.java
+++ b/plugins/google-genai/src/main/java/com/google/genkit/plugins/googlegenai/TtsModel.java
@@ -77,7 +77,10 @@ public class TtsModel implements Model {
   private static final Logger logger = LoggerFactory.getLogger(TtsModel.class);
 
   private static final Set SUPPORTED_TTS_MODELS =
-      Set.of("gemini-2.5-flash-preview-tts", "gemini-2.5-pro-preview-tts");
+      Set.of(
+          "gemini-3.1-flash-tts-preview",
+          "gemini-2.5-flash-preview-tts",
+          "gemini-2.5-pro-preview-tts");
 
   private final String modelName;
   private final GoogleGenAIPluginOptions options;
@@ -183,7 +186,7 @@ public ModelResponse run(
   }
 
   private ModelResponse callTts(ModelRequest request) throws Exception {
-    String prompt = extractPrompt(request);
+    String prompt = framePrompt(extractPrompt(request), request.getConfig());
     GenerateContentConfig config = buildConfig(request);
 
     logger.debug("Calling TTS model {} with prompt length: {}", modelName, prompt.length());
@@ -214,6 +217,33 @@ private String extractPrompt(ModelRequest request) {
     return prompt.toString();
   }
 
+  /**
+   * Frames the transcript with a clear synthesis preamble so the TTS model treats the input as text
+   * to speak rather than a request to answer.
+   *
+   * 
Gemini TTS models reject "vague" prompts with a 400 error ({@code "Model tried to generate
+   * text, but it should only be used for TTS"}). Google's guidance is to add a preamble instructing
+   * the model to synthesize speech and to explicitly label where the transcript begins.
+   *
+   * 
Callers that frame the prompt themselves (for example {@code "Say cheerfully: ..."}) can set
+   * a {@code ttsInstruction} config value: a custom preamble string, or an empty string to send the
+   * prompt verbatim with no framing.
+   *
+   * @param text the transcript to speak
+   * @param config the request config (may be null)
+   * @return the framed prompt
+   */
+  static String framePrompt(String text, Map config) {
+    if (config != null && config.containsKey("ttsInstruction")) {
+      Object instruction = config.get("ttsInstruction");
+      String instr = instruction != null ? instruction.toString() : "";
+      return instr.isEmpty() ? text : instr + "\n\n" + text;
+    }
+    return "Read the following transcript aloud verbatim, generating only speech audio. Do not"
+        + " respond to it or add any commentary.\n\nTranscript:\n"
+        + text;
+  }
+
   private GenerateContentConfig buildConfig(ModelRequest request) {
     GenerateContentConfig.Builder configBuilder = GenerateContentConfig.builder();
 
diff --git a/plugins/google-genai/src/main/java/com/google/genkit/plugins/googlegenai/VeoModel.java b/plugins/google-genai/src/main/java/com/google/genkit/plugins/googlegenai/VeoModel.java
index 2d997d73c..94be4c27c 100644
--- a/plugins/google-genai/src/main/java/com/google/genkit/plugins/googlegenai/VeoModel.java
+++ b/plugins/google-genai/src/main/java/com/google/genkit/plugins/googlegenai/VeoModel.java
@@ -61,11 +61,9 @@
  * 
Supported models:
  *
  * 
    
- *   
  • veo-2.0-generate-001
- *   
  • veo-3.0-generate-001
- *   
  • veo-3.0-fast-generate-001
  *   
  • veo-3.1-generate-preview
  *   
  • veo-3.1-fast-generate-preview
+ *   
  • veo-3.1-lite-generate-preview
  * 

  *
  * 
Configuration options (via custom config):
@@ -90,11 +88,9 @@ public class VeoModel implements Model {
 
   private static final Set SUPPORTED_VEO_MODELS =
       Set.of(
-          "veo-2.0-generate-001",
-          "veo-3.0-generate-001",
-          "veo-3.0-fast-generate-001",
           "veo-3.1-generate-preview",
-          "veo-3.1-fast-generate-preview");
+          "veo-3.1-fast-generate-preview",
+          "veo-3.1-lite-generate-preview");
 
   private static final long DEFAULT_POLL_INTERVAL_MS = 5000;
   private static final long DEFAULT_TIMEOUT_MS = 300000; // 5 minutes
diff --git a/plugins/google-genai/src/test/java/com/google/genkit/plugins/googlegenai/GoogleGenAIPluginTest.java b/plugins/google-genai/src/test/java/com/google/genkit/plugins/googlegenai/GoogleGenAIPluginTest.java
index 3748e1407..d41fb0bf2 100644
--- a/plugins/google-genai/src/test/java/com/google/genkit/plugins/googlegenai/GoogleGenAIPluginTest.java
+++ b/plugins/google-genai/src/test/java/com/google/genkit/plugins/googlegenai/GoogleGenAIPluginTest.java
@@ -80,6 +80,27 @@ void testSupportedModels() {
     assertTrue(GoogleGenAIPlugin.SUPPORTED_MODELS.size() > 0);
   }
 
+  @Test
+  void testSupportedOmniModels() {
+    assertNotNull(GoogleGenAIPlugin.SUPPORTED_OMNI_MODELS);
+    assertTrue(GoogleGenAIPlugin.SUPPORTED_OMNI_MODELS.contains("gemini-omni-flash-preview"));
+  }
+
+  @Test
+  void testSupportedTtsModelsIncludesLatest() {
+    assertTrue(GoogleGenAIPlugin.SUPPORTED_TTS_MODELS.contains("gemini-3.1-flash-tts-preview"));
+  }
+
+  @Test
+  void testRegistersOmniModel() {
+    GoogleGenAIPlugin plugin =
+        new GoogleGenAIPlugin(GoogleGenAIPluginOptions.builder().apiKey("test-key").build());
+    List> actions = plugin.init();
+    assertTrue(
+        actions.stream().anyMatch(a -> "googleai/gemini-omni-flash-preview".equals(a.getName())),
+        "Should register the Gemini Omni model");
+  }
+
   @Test
   void testGetOptions() {
     GoogleGenAIPluginOptions options =
diff --git a/plugins/google-genai/src/test/java/com/google/genkit/plugins/googlegenai/TtsModelTest.java b/plugins/google-genai/src/test/java/com/google/genkit/plugins/googlegenai/TtsModelTest.java
new file mode 100644
index 000000000..4f88fb0b8
--- /dev/null
+++ b/plugins/google-genai/src/test/java/com/google/genkit/plugins/googlegenai/TtsModelTest.java
@@ -0,0 +1,52 @@
+/*
+ * Copyright 2025 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package com.google.genkit.plugins.googlegenai;
+
+import static org.junit.jupiter.api.Assertions.assertEquals;
+import static org.junit.jupiter.api.Assertions.assertNotEquals;
+import static org.junit.jupiter.api.Assertions.assertTrue;
+
+import java.util.Map;
+import org.junit.jupiter.api.Test;
+
+/** Tests for {@link TtsModel} prompt framing. */
+class TtsModelTest {
+
+  @Test
+  void framePromptAddsPreambleForPlainText() {
+    String framed = TtsModel.framePrompt("Have a wonderful day", null);
+    assertTrue(framed.contains("Have a wonderful day"), "Should keep the transcript");
+    assertTrue(framed.toLowerCase().contains("transcript"), "Should label the transcript");
+    assertNotEquals(
+        "Have a wonderful day", framed, "Plain text should be framed with a synthesis preamble");
+  }
+
+  @Test
+  void framePromptRespectsEmptyInstructionOverride() {
+    String framed = TtsModel.framePrompt("Say cheerfully: Hi", Map.of("ttsInstruction", ""));
+    assertEquals(
+        "Say cheerfully: Hi", framed, "Empty ttsInstruction should send the prompt verbatim");
+  }
+
+  @Test
+  void framePromptUsesCustomInstruction() {
+    String framed = TtsModel.framePrompt("Hello", Map.of("ttsInstruction", "Say slowly:"));
+    assertEquals("Say slowly:\n\nHello", framed);
+  }
+}
diff --git a/plugins/groq/README.md b/plugins/groq/README.md
index 2ed30fe5d..39edcb810 100644
--- a/plugins/groq/README.md
+++ b/plugins/groq/README.md
@@ -19,8 +19,9 @@ This plugin provides integration with Groq's ultra-fast LLM inference.
 - `openai/gpt-oss-120b` - OpenAI GPT-OSS 120B with reasoning (~500 tokens/sec)
 - `openai/gpt-oss-20b` - OpenAI GPT-OSS 20B (~1000 tokens/sec)
 
-### Content Moderation
-- `meta-llama/llama-guard-4-12b` - Content moderation model (~1200 tokens/sec)
+### Agentic Systems
+- `groq/compound` - Agentic system with built-in web search + code execution
+- `groq/compound-mini` - Lightweight agentic system
 
 ## Using Custom Models
 
diff --git a/plugins/groq/src/main/java/com/google/genkit/plugins/groq/GroqPlugin.java b/plugins/groq/src/main/java/com/google/genkit/plugins/groq/GroqPlugin.java
index 3b59469df..4f355240f 100644
--- a/plugins/groq/src/main/java/com/google/genkit/plugins/groq/GroqPlugin.java
+++ b/plugins/groq/src/main/java/com/google/genkit/plugins/groq/GroqPlugin.java
@@ -43,11 +43,14 @@ public class GroqPlugin implements Plugin {
           // Meta Llama models
           "llama-3.1-8b-instant",
           "llama-3.3-70b-versatile",
-          "meta-llama/llama-guard-4-12b",
 
           // OpenAI GPT-OSS models
           "openai/gpt-oss-120b",
-          "openai/gpt-oss-20b");
+          "openai/gpt-oss-20b",
+
+          // Groq agentic systems (built-in web search + code execution)
+          "groq/compound",
+          "groq/compound-mini");
 
   private final CompatOAIPluginOptions options;
   private final List customModels = new ArrayList<>();
diff --git a/plugins/mistral/README.md b/plugins/mistral/README.md
index ddf92fdba..6f06009ff 100644
--- a/plugins/mistral/README.md
+++ b/plugins/mistral/README.md
@@ -13,14 +13,21 @@ This plugin provides integration with Mistral AI models.
 
 ## Supported Models
 
-- `mistral-large-3-25-12` - Latest flagship multimodal model (256K context)
-- `mistral-medium-3-1-25-08` - Balanced performance (128K context)
-- `mistral-small-3-2-25-06` - Efficient and fast (128K context)
-- `ministral-3-3b-25-12` - Compact 3B model (128K context)
-- `ministral-3-8b-25-12` - Balanced 8B model (128K context)
-- `ministral-3-14b-25-12` - Advanced 14B model (128K context)
-- `codestral-25-08` - Code generation specialist (256K context)
-- `devstral-2-25-12` - Developer-focused model
+- `mistral-large-2512` - Flagship model (256K context)
+- `mistral-medium-2604` - Mistral Medium 3.5 (128K context)
+- `mistral-small-2603` - Mistral Small 4 (128K context)
+- `magistral-medium-2509`, `magistral-small-2509` - Reasoning models
+- `ministral-3b-2512`, `ministral-8b-2512`, `ministral-14b-2512` - Compact models
+- `codestral-2508` - Code generation specialist (256K context)
+- `devstral-2512` - Developer/agentic coding model
+- `open-mistral-nemo` - Open-source multilingual model
+
+## Embeddings
+
+- `mistral-embed`
+- `codestral-embed`
+
+Register additional embedding models with `customEmbeddingModel(...)`.
 
 ## Using Custom Models
 
diff --git a/plugins/mistral/src/main/java/com/google/genkit/plugins/mistral/MistralPlugin.java b/plugins/mistral/src/main/java/com/google/genkit/plugins/mistral/MistralPlugin.java
index 03bb33669..6d4ecd838 100644
--- a/plugins/mistral/src/main/java/com/google/genkit/plugins/mistral/MistralPlugin.java
+++ b/plugins/mistral/src/main/java/com/google/genkit/plugins/mistral/MistralPlugin.java
@@ -20,6 +20,7 @@
 
 import com.google.genkit.core.Action;
 import com.google.genkit.core.Plugin;
+import com.google.genkit.plugins.compatoai.CompatOAIEmbedder;
 import com.google.genkit.plugins.compatoai.CompatOAIModel;
 import com.google.genkit.plugins.compatoai.CompatOAIPluginOptions;
 import java.util.ArrayList;
@@ -42,7 +43,9 @@ public class MistralPlugin implements Plugin {
       Arrays.asList(
           // Flagship models
           "mistral-large-2512",
+          "mistral-medium-2604",
           "mistral-medium-2508",
+          "mistral-small-2603",
           "mistral-small-2506",
 
           // Reasoning models
@@ -54,9 +57,6 @@ public class MistralPlugin implements Plugin {
           "ministral-8b-2512",
           "ministral-14b-2512",
 
-          // Vision models
-          "pixtral-large-2411",
-
           // Code models
           "codestral-2508",
           "devstral-2512",
@@ -64,8 +64,13 @@ public class MistralPlugin implements Plugin {
           // Open source
           "open-mistral-nemo");
 
+  /** Supported Mistral embedding models. */
+  public static final List SUPPORTED_EMBEDDING_MODELS =
+      Arrays.asList("mistral-embed", "codestral-embed");
+
   private final CompatOAIPluginOptions options;
   private final List customModels = new ArrayList<>();
+  private final List customEmbeddingModels = new ArrayList<>();
 
   /** Creates a MistralPlugin with default options (using MISTRAL_API_KEY environment variable). */
   public MistralPlugin() {
@@ -148,8 +153,26 @@ public String getName() {
       logger.debug("Created custom Mistral model: {}", modelName);
     }
 
+    // Register Mistral embedding models
+    for (String modelName : SUPPORTED_EMBEDDING_MODELS) {
+      CompatOAIEmbedder embedder =
+          new CompatOAIEmbedder("mistral/" + modelName, modelName, "Mistral " + modelName, options);
+      actions.add(embedder);
+      logger.debug("Created Mistral embedder: {}", modelName);
+    }
+
+    // Register custom embedding models added via customEmbeddingModel()
+    for (String modelName : customEmbeddingModels) {
+      CompatOAIEmbedder embedder =
+          new CompatOAIEmbedder("mistral/" + modelName, modelName, "Mistral " + modelName, options);
+      actions.add(embedder);
+      logger.debug("Created custom Mistral embedder: {}", modelName);
+    }
+
     logger.info(
-        "Mistral plugin initialized with {} models", SUPPORTED_MODELS.size() + customModels.size());
+        "Mistral plugin initialized with {} models and {} embedders",
+        SUPPORTED_MODELS.size() + customModels.size(),
+        SUPPORTED_EMBEDDING_MODELS.size() + customEmbeddingModels.size());
 
     return actions;
   }
@@ -167,6 +190,19 @@ public MistralPlugin customModel(String modelName) {
     return this;
   }
 
+  /**
+   * Registers a custom embedding model name. Use this to work with embedding models not in the
+   * default list. Call this method before passing the plugin to Genkit.builder().
+   *
+   * @param modelName the embedding model name (e.g., "mistral-embed-2601")
+   * @return this plugin instance for method chaining
+   */
+  public MistralPlugin customEmbeddingModel(String modelName) {
+    customEmbeddingModels.add(modelName);
+    logger.debug("Added custom embedding model to be registered: {}", modelName);
+    return this;
+  }
+
   /**
    * Gets the plugin options.
    *
diff --git a/plugins/mistral/src/test/java/com/google/genkit/plugins/mistral/MistralPluginTest.java b/plugins/mistral/src/test/java/com/google/genkit/plugins/mistral/MistralPluginTest.java
index 01cc3ef9e..d7feab971 100644
--- a/plugins/mistral/src/test/java/com/google/genkit/plugins/mistral/MistralPluginTest.java
+++ b/plugins/mistral/src/test/java/com/google/genkit/plugins/mistral/MistralPluginTest.java
@@ -92,6 +92,26 @@ void testSupportedModels() {
     assertTrue(MistralPlugin.SUPPORTED_MODELS.contains("mistral-small-2506"));
   }
 
+  @Test
+  void testSupportedEmbeddingModels() {
+    assertNotNull(MistralPlugin.SUPPORTED_EMBEDDING_MODELS);
+    assertTrue(MistralPlugin.SUPPORTED_EMBEDDING_MODELS.contains("mistral-embed"));
+  }
+
+  @Test
+  void testRegistersEmbedders() {
+    MistralPlugin plugin =
+        new MistralPlugin(
+            CompatOAIPluginOptions.builder()
+                .apiKey("test-key")
+                .baseUrl("https://api.mistral.ai/v1")
+                .build());
+    List> actions = plugin.init();
+    assertTrue(
+        actions.stream().anyMatch(a -> "mistral/mistral-embed".equals(a.getName())),
+        "Should register the mistral-embed embedder");
+  }
+
   @Test
   void testCustomModel() {
     MistralPlugin plugin =
diff --git a/plugins/openai/README.md b/plugins/openai/README.md
index 07e329a03..656ef953b 100644
--- a/plugins/openai/README.md
+++ b/plugins/openai/README.md
@@ -93,11 +93,12 @@ Genkit genkit = Genkit.builder()
 ## Supported Models
 
 ### Chat Models
-- `gpt-5.2`, `gpt-5.1`, `gpt-5`
-- `gpt-4o`, `gpt-4o-mini`
-- `gpt-4-turbo`, `gpt-4`, `gpt-4-32k`
-- `gpt-3.5-turbo`, `gpt-3.5-turbo-16k`
-- `o1-preview`, `o1-mini`
+- `gpt-5.5`, `gpt-5.5-pro`
+- `gpt-5.4`, `gpt-5.4-pro`, `gpt-5.4-mini`, `gpt-5.4-nano`
+- `gpt-5.2`, `gpt-5.1`, `gpt-5`, `gpt-5-mini`, `gpt-5-nano`, `gpt-5-pro`
+- `o3`, `o3-pro`
+- `gpt-4.1`, `gpt-4.1-mini`
+- `gpt-4o`, `gpt-4o-mini`, `gpt-4-turbo`, `gpt-4`, `gpt-3.5-turbo`
 
 ### Embedding Models
 - `text-embedding-3-small`
diff --git a/plugins/openai/src/main/java/com/google/genkit/plugins/openai/OpenAIPlugin.java b/plugins/openai/src/main/java/com/google/genkit/plugins/openai/OpenAIPlugin.java
index 27a096dce..e75faff45 100644
--- a/plugins/openai/src/main/java/com/google/genkit/plugins/openai/OpenAIPlugin.java
+++ b/plugins/openai/src/main/java/com/google/genkit/plugins/openai/OpenAIPlugin.java
@@ -42,19 +42,33 @@ public class OpenAIPlugin implements Plugin {
   /** Supported GPT models. */
   public static final List SUPPORTED_MODELS =
       Arrays.asList(
+          // GPT-5.5 / 5.4 family
+          "gpt-5.5",
+          "gpt-5.5-pro",
+          "gpt-5.4",
+          "gpt-5.4-pro",
+          "gpt-5.4-mini",
+          "gpt-5.4-nano",
+          // GPT-5.2 / 5.1 / 5 family
           "gpt-5.2",
+          "gpt-5.2-pro",
           "gpt-5.1",
           "gpt-5",
+          "gpt-5-mini",
+          "gpt-5-nano",
+          "gpt-5-pro",
+          // o-series reasoning models
+          "o3",
+          "o3-pro",
+          // GPT-4.1 / 4o / 4
+          "gpt-4.1",
+          "gpt-4.1-mini",
           "gpt-4o",
           "gpt-4o-mini",
           "gpt-4-turbo",
           "gpt-4-turbo-preview",
           "gpt-4",
-          "gpt-4-32k",
-          "gpt-3.5-turbo",
-          "gpt-3.5-turbo-16k",
-          "o1-preview",
-          "o1-mini");
+          "gpt-3.5-turbo");
 
   /** Supported embedding models. */
   public static final List SUPPORTED_EMBEDDING_MODELS =
@@ -62,7 +76,7 @@ public class OpenAIPlugin implements Plugin {
 
   /** Supported image generation models. */
   public static final List SUPPORTED_IMAGE_MODELS =
-      Arrays.asList("dall-e-3", "dall-e-2", "gpt-image-1");
+      Arrays.asList("gpt-image-2", "gpt-image-1.5", "gpt-image-1", "dall-e-3");
 
   private final OpenAIPluginOptions legacyOptions;
   private final CompatOAIPluginOptions compatOptions;
diff --git a/plugins/openai/src/test/java/com/google/genkit/plugins/openai/OpenAIPluginTest.java b/plugins/openai/src/test/java/com/google/genkit/plugins/openai/OpenAIPluginTest.java
index 843c85b07..e550ea3ca 100644
--- a/plugins/openai/src/test/java/com/google/genkit/plugins/openai/OpenAIPluginTest.java
+++ b/plugins/openai/src/test/java/com/google/genkit/plugins/openai/OpenAIPluginTest.java
@@ -97,7 +97,7 @@ void testSupportedEmbeddingModels() {
   void testSupportedImageModels() {
     assertNotNull(OpenAIPlugin.SUPPORTED_IMAGE_MODELS);
     assertTrue(OpenAIPlugin.SUPPORTED_IMAGE_MODELS.contains("dall-e-3"));
-    assertTrue(OpenAIPlugin.SUPPORTED_IMAGE_MODELS.contains("dall-e-2"));
+    assertTrue(OpenAIPlugin.SUPPORTED_IMAGE_MODELS.contains("gpt-image-2"));
     assertTrue(OpenAIPlugin.SUPPORTED_IMAGE_MODELS.contains("gpt-image-1"));
   }
 
diff --git a/plugins/xai/README.md b/plugins/xai/README.md
index 6c7d617e5..65bafbaaa 100644
--- a/plugins/xai/README.md
+++ b/plugins/xai/README.md
@@ -13,10 +13,11 @@ This plugin provides integration with XAI (x.ai / Grok) models.
 
 ## Supported Models
 
-- `grok-4` - Latest flagship model (131K context)
-- `grok-4-1-fast` - Optimized for agentic tool calling (2M context)
-- `grok-3` - Previous generation (131K context)
-- `grok-3-mini` - Efficient small model (131K context)
+- `grok-4.3` - Latest flagship model (2M context)
+- `grok-4.20-0309-reasoning` - Reasoning mode (2M context)
+- `grok-4.20-0309-non-reasoning` - Non-reasoning mode (2M context)
+- `grok-4.20-multi-agent-0309` - Multi-agent mode (2M context)
+- `grok-build-0.1` - Agentic coding model (256K context)
 
 ## Using Custom Models
 
@@ -56,7 +57,7 @@ Genkit genkit = Genkit.builder()
 // Use the model
 ModelResponse response = genkit.generate(
     GenerateOptions.builder()
-        .model("xai/grok-4")
+        .model("xai/grok-4.3")
         .prompt("Tell me a joke")
         .build()
 );
diff --git a/plugins/xai/src/main/java/com/google/genkit/plugins/xai/XAIPlugin.java b/plugins/xai/src/main/java/com/google/genkit/plugins/xai/XAIPlugin.java
index e2b0b781c..908c7a7c3 100644
--- a/plugins/xai/src/main/java/com/google/genkit/plugins/xai/XAIPlugin.java
+++ b/plugins/xai/src/main/java/com/google/genkit/plugins/xai/XAIPlugin.java
@@ -40,23 +40,16 @@ public class XAIPlugin implements Plugin {
   /** Supported XAI models. */
   public static final List SUPPORTED_MODELS =
       Arrays.asList(
-          // Latest flagship models
-          "grok-4",
-          "grok-4-1-fast",
-
-          // Reasoning variants
-          "grok-4-1-fast-reasoning",
-          "grok-4-1-fast-non-reasoning",
-          "grok-4-fast-reasoning",
-          "grok-4-fast-non-reasoning",
-
-          // Code model
-          "grok-code-fast-1",
-
-          // Previous generation
-          "grok-4-0709",
-          "grok-3",
-          "grok-3-mini");
+          // Latest flagship model
+          "grok-4.3",
+
+          // Grok 4.20 variants
+          "grok-4.20-0309-reasoning",
+          "grok-4.20-0309-non-reasoning",
+          "grok-4.20-multi-agent-0309",
+
+          // Agentic coding model
+          "grok-build-0.1");
 
   private final CompatOAIPluginOptions options;
   private final List customModels = new ArrayList<>();
diff --git a/plugins/xai/src/test/java/com/google/genkit/plugins/xai/XAIPluginTest.java b/plugins/xai/src/test/java/com/google/genkit/plugins/xai/XAIPluginTest.java
index 4b7fe0334..d656fea8c 100644
--- a/plugins/xai/src/test/java/com/google/genkit/plugins/xai/XAIPluginTest.java
+++ b/plugins/xai/src/test/java/com/google/genkit/plugins/xai/XAIPluginTest.java
@@ -67,10 +67,9 @@ void testCreateWithApiKey() {
   @Test
   void testSupportedModels() {
     assertNotNull(XAIPlugin.SUPPORTED_MODELS);
-    assertTrue(XAIPlugin.SUPPORTED_MODELS.contains("grok-4"));
-    assertTrue(XAIPlugin.SUPPORTED_MODELS.contains("grok-4-1-fast"));
-    assertTrue(XAIPlugin.SUPPORTED_MODELS.contains("grok-3"));
-    assertTrue(XAIPlugin.SUPPORTED_MODELS.contains("grok-code-fast-1"));
+    assertTrue(XAIPlugin.SUPPORTED_MODELS.contains("grok-4.3"));
+    assertTrue(XAIPlugin.SUPPORTED_MODELS.contains("grok-4.20-0309-reasoning"));
+    assertTrue(XAIPlugin.SUPPORTED_MODELS.contains("grok-build-0.1"));
   }
 
   @Test
diff --git a/samples/anthropic/README.md b/samples/anthropic/README.md
index afc780eb0..d0852be85 100644
--- a/samples/anthropic/README.md
+++ b/samples/anthropic/README.md
@@ -21,17 +21,19 @@ The Anthropic plugin supports the following Claude models:
 - `claude-sonnet-4-5-20250929` - Excellent balance of capability and speed
 - `claude-haiku-4-5-20251001` - Fast and efficient
 
+### Claude 5 Family
+- `claude-fable-5` - Most capable widely released model
+- `claude-sonnet-5` - Balanced speed and intelligence
+
 ### Claude 4 Family
+- `claude-opus-4-8` - Flagship Opus
+- `claude-opus-4-7` - Claude Opus 4.7
+- `claude-opus-4-6` - Claude Opus 4.6
+- `claude-sonnet-4-6` - Claude Sonnet 4.6
 - `claude-opus-4-1-20250805` - Claude 4.1 Opus
 - `claude-opus-4-20250514` - Claude 4 Opus
 - `claude-sonnet-4-20250514` - Claude 4 Sonnet
 
-### Claude 3 Family
-- `claude-3-7-sonnet-20250219` - Claude 3.7 Sonnet
-- `claude-3-5-haiku-20241022` - Claude 3.5 Haiku
-- `claude-3-opus-20240229` - Most powerful Claude 3 model
-- `claude-3-haiku-20240307` - Fastest Claude 3 model
-
 ## Prerequisites
 
 - Java 21+
diff --git a/samples/azure-foundry/src/main/java/com/google/genkit/samples/AzureFoundrySample.java b/samples/azure-foundry/src/main/java/com/google/genkit/samples/AzureFoundrySample.java
index d3f382a39..4cf381f1f 100644
--- a/samples/azure-foundry/src/main/java/com/google/genkit/samples/AzureFoundrySample.java
+++ b/samples/azure-foundry/src/main/java/com/google/genkit/samples/AzureFoundrySample.java
@@ -38,9 +38,9 @@
  * streaming for real-time responses - Use Azure Managed Identity for authentication - Expose flows
  * via HTTP endpoints
  *
- * 
Supported models include: - Azure OpenAI: gpt-5-turbo, o1, o3-mini, gpt-4o, gpt-4o-mini,
- * gpt-4, gpt-35-turbo - Azure Direct: MAI-DS-R1, Grok-4, Llama-3.3, DeepSeek-V3/R1, GPT-OSS -
- * Partner: Claude Opus/Sonnet/Haiku 4.x
+ * 
Supported models include: - Azure OpenAI: gpt-5.5, gpt-5.4, gpt-5, o3, gpt-4.1, gpt-4o - Azure
+ * Direct: Grok-4, Grok-4.1-fast, Llama-3.3, DeepSeek-V3.2, Mistral-Large-3, GPT-OSS - Partner:
+ * Claude Opus/Sonnet 4.x and 5
  *
  * 
To run: 1. Set AZURE_AI_FOUNDRY_ENDPOINT environment variable 2. Configure authentication (API
  * key or Azure credentials) 3. Ensure models are deployed in your Azure AI Foundry project 4. Run:
diff --git a/samples/firebase/src/main/java/com/google/genkit/samples/firebase/FirestoreRAGSample.java b/samples/firebase/src/main/java/com/google/genkit/samples/firebase/FirestoreRAGSample.java
index 1992fdd6b..bdea3ee2c 100644
--- a/samples/firebase/src/main/java/com/google/genkit/samples/firebase/FirestoreRAGSample.java
+++ b/samples/firebase/src/main/java/com/google/genkit/samples/firebase/FirestoreRAGSample.java
@@ -129,7 +129,7 @@ public static void main(String[] args) {
                                 true) // Auto-create database if it doesn't exist
                             .createVectorIndexIfNotExists(true) // Auto-create vector index if it
                             // doesn't exist
-                            .embedderDimension(768) // text-embedding-004 outputs 768 dimensions
+                            .embedderDimension(768) // gemini-embedding-001 outputs 768 dimensions
                             .build())
                     .build())
             .plugin(jetty)
diff --git a/samples/google-genai/README.md b/samples/google-genai/README.md
index 949330d1c..567677d77 100644
--- a/samples/google-genai/README.md
+++ b/samples/google-genai/README.md
@@ -119,9 +119,9 @@ The Google GenAI plugin provides access to:
 | Model | Description |
 |-------|-------------|
 | `googleai/gemini-2.5-flash` | Fast, efficient Gemini model |
-| `googleai/gemini-1.5-pro` | Advanced reasoning capabilities |
-| `googleai/gemini-1.5-flash` | Balanced speed and capability |
-| `googleai/imagen-3.0-generate-002` | Image generation |
+| `googleai/gemini-3.1-pro-preview` | Advanced reasoning capabilities |
+| `googleai/gemini-3.5-flash` | Balanced speed and capability |
+| `googleai/imagen-4.0-fast-generate-001` | Image generation |
 | `googleai/gemini-embedding-001` | Text embeddings |
 
 ## Code Highlights
@@ -166,7 +166,7 @@ genkit.defineFlow("imageGeneration", String.class, String.class,
     (ctx, prompt) -> {
         ModelResponse response = genkit.generate(
             GenerateOptions.builder()
-                .model("googleai/imagen-3.0-generate-002")
+                .model("googleai/imagen-4.0-fast-generate-001")
                 .prompt(prompt)
                 .build());
         // Save generated image to file
diff --git a/samples/google-genai/src/main/java/com/google/genkit/samples/GoogleGenAIApp.java b/samples/google-genai/src/main/java/com/google/genkit/samples/GoogleGenAIApp.java
index 4558b5d86..c572ecc38 100644
--- a/samples/google-genai/src/main/java/com/google/genkit/samples/GoogleGenAIApp.java
+++ b/samples/google-genai/src/main/java/com/google/genkit/samples/GoogleGenAIApp.java
@@ -84,6 +84,7 @@ public static void main(String[] args) throws Exception {
     defineImageGenerationFlow(genkit);
     defineTextToSpeechFlow(genkit);
     defineVideoGenerationFlow(genkit);
+    defineOmniFlow(genkit);
 
     System.out.println("Server started on http://localhost:8080");
     System.out.println("Use Genkit Developer UI at http://localhost:4000 to interact with flows");
@@ -96,6 +97,10 @@ public static void main(String[] args) throws Exception {
     System.out.println("  - textToSpeech: Generate audio with TTS (saves to " + OUTPUT_DIR + "/)");
     System.out.println(
         "  - videoGeneration: Generate videos with Veo (saves to " + OUTPUT_DIR + "/)");
+    System.out.println(
+        "  - omniVideo: Generate + conversationally edit video with Gemini Omni (saves to "
+            + OUTPUT_DIR
+            + "/)");
     System.out.println(
         "\nGenerated media files will be saved to: " + new File(OUTPUT_DIR).getAbsolutePath());
     System.out.println("\nPress Ctrl+C to stop the server.");
@@ -401,7 +406,7 @@ private static void defineVideoGenerationFlow(Genkit genkit) {
           ModelResponse response =
               genkit.generate(
                   GenerateOptions.builder()
-                      .model("googleai/veo-3.0-generate-001")
+                      .model("googleai/veo-3.1-generate-preview")
                       .prompt(prompt)
                       .config(config)
                       .build());
@@ -460,4 +465,88 @@ private static void defineVideoGenerationFlow(Genkit genkit) {
           return "No videos generated";
         });
   }
+
+  /**
+   * Demonstrates Gemini Omni: generate a video, then conversationally edit it by passing the
+   * returned interaction id back as {@code previousInteractionId}.
+   */
+  private static void defineOmniFlow(Genkit genkit) {
+    genkit.defineFlow(
+        "omniVideo",
+        String.class,
+        String.class,
+        (ctx, prompt) -> {
+          StringBuilder result = new StringBuilder();
+
+          // Turn 1: generate a video from the prompt.
+          ModelResponse first =
+              genkit.generate(
+                  GenerateOptions.builder()
+                      .model("googleai/gemini-omni-flash-preview")
+                      .prompt(prompt)
+                      .config(
+                          GenerationConfig.builder()
+                              .custom(Map.of("aspectRatio", "16:9", "duration", "10s"))
+                              .build())
+                      .build());
+          result.append("--- Turn 1 (generate) ---\n");
+          result.append(saveVideosFromResponse(first, "omni"));
+
+          String interactionId =
+              first.getCustom() != null ? (String) first.getCustom().get("interactionId") : null;
+          result.append("Interaction id: ").append(interactionId).append("\n");
+
+          // Turn 2: conversational edit that builds on the previous result.
+          if (interactionId != null) {
+            ModelResponse edited =
+                genkit.generate(
+                    GenerateOptions.builder()
+                        .model("googleai/gemini-omni-flash-preview")
+                        .prompt("Brighten the background and add a slow push-in on the subject.")
+                        .config(
+                            GenerationConfig.builder()
+                                .custom(Map.of("previousInteractionId", interactionId))
+                                .build())
+                        .build());
+            result.append("\n--- Turn 2 (edit) ---\n");
+            result.append(saveVideosFromResponse(edited, "omni_edited"));
+          }
+
+          return result.toString();
+        });
+  }
+
+  /** Saves any video media parts in a model response to disk, returning a summary string. */
+  private static String saveVideosFromResponse(ModelResponse response, String prefix) {
+    if (response.getMessage() == null || response.getMessage().getContent() == null) {
+      return "No video returned.\n";
+    }
+    StringBuilder result = new StringBuilder();
+    int count = 0;
+    for (Part part : response.getMessage().getContent()) {
+      if (part.getMedia() == null) {
+        continue;
+      }
+      count++;
+      String url = part.getMedia().getUrl();
+      if (url.startsWith("data:")) {
+        String base64Data = extractBase64FromDataUrl(url);
+        String filename = prefix + "_" + System.currentTimeMillis() + "_" + count + ".mp4";
+        try {
+          result
+              .append("Video saved to: ")
+              .append(saveBase64ToFile(base64Data, filename))
+              .append("\n");
+        } catch (IOException e) {
+          result.append("Video failed to save: ").append(e.getMessage()).append("\n");
+        }
+      } else {
+        result.append("Video available at: ").append(url).append("\n");
+      }
+    }
+    if (count == 0) {
+      result.append("No video parts in response.\n");
+    }
+    return result.toString();
+  }
 }
diff --git a/samples/groq/README.md b/samples/groq/README.md
index 00c2a57cb..a4b7a6836 100644
--- a/samples/groq/README.md
+++ b/samples/groq/README.md
@@ -20,7 +20,8 @@ This sample demonstrates integration with Groq's ultra-fast LLM inference using
 - `llama-3.3-70b-versatile` - Latest Meta Llama 3.3 70B (~280 tokens/sec)
 - `openai/gpt-oss-120b` - OpenAI GPT-OSS 120B with reasoning (~500 tokens/sec)
 - `openai/gpt-oss-20b` - OpenAI GPT-OSS 20B (~1000 tokens/sec)
-- `meta-llama/llama-guard-4-12b` - Content moderation (~1200 tokens/sec)
+- `groq/compound` - Agentic system (web search + code execution)
+- `groq/compound-mini` - Lightweight agentic system
 
 ## Prerequisites
 
@@ -67,9 +68,9 @@ The Dev UI will be available at http://localhost:4000
 | `chat` | llama-3.3-70b-versatile | Chat with most capable model |
 | `timeAssistant` | llama-3.3-70b-versatile | Time zone assistant with tool |
 | `fastStreaming` | llama-3.1-8b-instant | Ultra-fast streaming responses |
-| `qualityChat` | mixtral-8x7b-32768 | High-quality chat with large context |
-| `efficientChat` | gemma2-9b-it | Efficient chat |
-| `realTimeQA` | llama-3.1-70b-versatile | Real-time Q&A with timing |
+| `qualityChat` | openai/gpt-oss-120b | High-quality chat |
+| `efficientChat` | openai/gpt-oss-20b | Efficient chat |
+| `realTimeQA` | llama-3.3-70b-versatile | Real-time Q&A with timing |
 | `speedComparison` | multiple | Benchmark Groq's speed |
 
 ## Example API Calls
diff --git a/samples/groq/src/main/java/com/google/genkit/samples/GroqSample.java b/samples/groq/src/main/java/com/google/genkit/samples/GroqSample.java
index 2a9abf70d..80f217c9b 100644
--- a/samples/groq/src/main/java/com/google/genkit/samples/GroqSample.java
+++ b/samples/groq/src/main/java/com/google/genkit/samples/GroqSample.java
@@ -174,7 +174,7 @@ public static void main(String[] args) throws Exception {
               ModelResponse response =
                   genkit.generate(
                       GenerateOptions.builder()
-                          .model("groq/mixtral-8x7b-32768")
+                          .model("groq/openai/gpt-oss-120b")
                           .system(
                               "You are a knowledgeable assistant providing detailed, high-quality responses.")
                           .prompt(userMessage)
@@ -188,7 +188,7 @@ public static void main(String[] args) throws Exception {
               return response.getText();
             });
 
-    // Define an efficient chat flow using Gemma 2
+    // Define an efficient chat flow using GPT-OSS 20B
     Flow efficientChatFlow =
         genkit.defineFlow(
             "efficientChat",
@@ -198,7 +198,7 @@ public static void main(String[] args) throws Exception {
               ModelResponse response =
                   genkit.generate(
                       GenerateOptions.builder()
-                          .model("groq/gemma2-9b-it")
+                          .model("groq/openai/gpt-oss-20b")
                           .system("You are an efficient, helpful assistant.")
                           .prompt(userMessage)
                           .config(
@@ -223,7 +223,7 @@ public static void main(String[] args) throws Exception {
               ModelResponse response =
                   genkit.generate(
                       GenerateOptions.builder()
-                          .model("groq/llama-3.1-70b-versatile")
+                          .model("groq/llama-3.3-70b-versatile")
                           .system(
                               "You are a real-time Q&A assistant. Provide quick, accurate answers.")
                           .prompt(question)
@@ -288,8 +288,8 @@ public static void main(String[] args) throws Exception {
     System.out.println("  POST http://localhost:8080/api/flows/timeAssistant (uses tools)");
     System.out.println(
         "  POST http://localhost:8080/api/flows/fastStreaming (llama-3.1-8b ultra-fast)");
-    System.out.println("  POST http://localhost:8080/api/flows/qualityChat (mixtral-8x7b)");
-    System.out.println("  POST http://localhost:8080/api/flows/efficientChat (gemma2-9b)");
+    System.out.println("  POST http://localhost:8080/api/flows/qualityChat (gpt-oss-120b)");
+    System.out.println("  POST http://localhost:8080/api/flows/efficientChat (gpt-oss-20b)");
     System.out.println("  POST http://localhost:8080/api/flows/realTimeQA (real-time Q&A)");
     System.out.println(
         "  POST http://localhost:8080/api/flows/speedComparison (benchmark Groq speed)");
diff --git a/samples/mistral/README.md b/samples/mistral/README.md
index 1418dc865..abc418d6e 100644
--- a/samples/mistral/README.md
+++ b/samples/mistral/README.md
@@ -29,9 +29,6 @@ This sample demonstrates integration with Mistral AI models using Genkit Java.
 - `ministral-8b-2512` - Balanced 8B model (262K context)
 - `ministral-14b-2512` - Advanced 14B model (262K context)
 
-### Vision Models
-- `pixtral-large-2411` - Multimodal with vision (131K context)
-
 ### Code Models
 - `codestral-2508` - Code generation specialist (256K context)
 - `devstral-2512` - Code-agentic model (262K context)
diff --git a/samples/mistral/src/main/java/com/google/genkit/samples/MistralSample.java b/samples/mistral/src/main/java/com/google/genkit/samples/MistralSample.java
index 610bd5080..700b20dc6 100644
--- a/samples/mistral/src/main/java/com/google/genkit/samples/MistralSample.java
+++ b/samples/mistral/src/main/java/com/google/genkit/samples/MistralSample.java
@@ -283,7 +283,7 @@ public static void main(String[] args) throws Exception {
     System.out.println("  POST http://localhost:8080/api/flows/quickQA (ministral-3b)");
     System.out.println("  POST http://localhost:8080/api/flows/efficientChat (ministral-8b)");
     System.out.println(
-        "  POST http://localhost:8080/api/flows/creativeWriting (pixtral-large with streaming)");
+        "  POST http://localhost:8080/api/flows/creativeWriting (mistral-large with streaming)");
     System.out.println("");
     System.out.println("Example usage:");
     System.out.println(
diff --git a/samples/pinecone/README.md b/samples/pinecone/README.md
index 4dfdcca16..9da9cc878 100644
--- a/samples/pinecone/README.md
+++ b/samples/pinecone/README.md
@@ -168,7 +168,7 @@ Either create the index in the Pinecone console or enable auto-creation.
 Vector dimension does not match index dimension
 ```
 
-Ensure your index was created with 768 dimensions to match `text-embedding-004`.
+Ensure your index was created with 768 dimensions to match `gemini-embedding-001`.
 
 ### Rate limiting
 
diff --git a/samples/xai/README.md b/samples/xai/README.md
index 389c28692..4635301d6 100644
--- a/samples/xai/README.md
+++ b/samples/xai/README.md
@@ -10,27 +10,20 @@ This sample demonstrates integration with XAI (Grok) models using Genkit Java.
 - **Text Generation** - Generate text with latest Grok 4 models
 - **Streaming** - Real-time response streaming
 - **Code Generation** - Generate code with Grok 3
-- **Fast Tool Calling** - Optimized agentic workflows with Grok 4.1 Fast
+- **Fast Tool Calling** - Optimized agentic workflows with Grok 4.3
 
 ## Supported Models
 
 ### Latest Flagship
-- `grok-4` - Latest flagship model (131K context)
-- `grok-4-1-fast` - Optimized for agentic tool calling (2M context)
+- `grok-4.3` - Latest flagship model (2M context)
 
-### Reasoning Variants
-- `grok-4-1-fast-reasoning` - Fast reasoning mode (2M context)
-- `grok-4-1-fast-non-reasoning` - Fast without reasoning (2M context)
-- `grok-4-fast-reasoning` - Standard fast reasoning (2M context)
-- `grok-4-fast-non-reasoning` - Standard fast without reasoning (2M context)
+### Grok 4.20 Variants
+- `grok-4.20-0309-reasoning` - Reasoning mode (2M context)
+- `grok-4.20-0309-non-reasoning` - Non-reasoning mode (2M context)
+- `grok-4.20-multi-agent-0309` - Multi-agent mode (2M context)
 
-### Code Generation
-- `grok-code-fast-1` - Specialized for code generation (256K context)
-
-### Previous Generation
-- `grok-4-0709` - July 2024 version (256K context)
-- `grok-3` - Powerful previous generation (131K context)
-- `grok-3-mini` - Efficient small model (131K context)
+### Agentic Coding
+- `grok-build-0.1` - Specialized for agentic coding (256K context)
 
 ## Prerequisites
 
@@ -74,12 +67,12 @@ The Dev UI will be available at http://localhost:4000
 | Flow | Model | Description |
 |------|-------|-------------|
 | `greeting` | - | Simple greeting flow |
-| `chat` | grok-4 | Chat with latest Grok |
-| `weatherAssistant` | grok-4-1-fast | Fast tool calling for weather |
-| `streamingChat` | grok-4 | Streaming chat responses |
-| `generateCode` | grok-3 | Code generation |
-| `analyze` | grok-3-mini | Fast text analysis |
-| `creativeWriting` | grok-4 | Creative writing with streaming |
+| `chat` | grok-4.3 | Chat with latest Grok |
+| `weatherAssistant` | grok-4.3 | Tool calling for weather |
+| `streamingChat` | grok-4.3 | Streaming chat responses |
+| `generateCode` | grok-build-0.1 | Agentic code generation |
+| `analyze` | grok-4.20-0309-reasoning | Reasoning-based text analysis |
+| `creativeWriting` | grok-4.3 | Creative writing with streaming |
 
 ## Example API Calls
 
diff --git a/samples/xai/src/main/java/com/google/genkit/samples/XAISample.java b/samples/xai/src/main/java/com/google/genkit/samples/XAISample.java
index bdd1a7293..8a1fafd37 100644
--- a/samples/xai/src/main/java/com/google/genkit/samples/XAISample.java
+++ b/samples/xai/src/main/java/com/google/genkit/samples/XAISample.java
@@ -66,7 +66,7 @@ public static void main(String[] args) throws Exception {
               ModelResponse response =
                   genkit.generate(
                       GenerateOptions.builder()
-                          .model("xai/grok-4")
+                          .model("xai/grok-4.3")
                           .system("You are Grok, a witty and helpful AI assistant created by xAI.")
                           .prompt(userMessage)
                           .build());
@@ -107,7 +107,7 @@ public static void main(String[] args) throws Exception {
               ModelResponse response =
                   genkit.generate(
                       GenerateOptions.builder()
-                          .model("xai/grok-4-1-fast")
+                          .model("xai/grok-4.3")
                           .system(
                               "You are a helpful weather assistant. Use the getWeather tool to provide weather information.")
                           .prompt(userMessage)
@@ -133,7 +133,7 @@ public static void main(String[] args) throws Exception {
               ModelResponse response =
                   genkit.generateStream(
                       GenerateOptions.builder()
-                          .model("xai/grok-4")
+                          .model("xai/grok-4.3")
                           .system("You are Grok, providing detailed and engaging responses.")
                           .prompt(userMessage)
                           .config(GenerationConfig.builder().maxOutputTokens(1000).build())
@@ -160,7 +160,7 @@ public static void main(String[] args) throws Exception {
               ModelResponse response =
                   genkit.generate(
                       GenerateOptions.builder()
-                          .model("xai/grok-3")
+                          .model("xai/grok-build-0.1")
                           .system(
                               "You are an expert programmer. Write clean, well-documented code with helpful comments.")
                           .prompt(prompt)
@@ -184,7 +184,7 @@ public static void main(String[] args) throws Exception {
               ModelResponse response =
                   genkit.generate(
                       GenerateOptions.builder()
-                          .model("xai/grok-3-mini")
+                          .model("xai/grok-4.20-0309-reasoning")
                           .system(
                               "You are an analytical expert. Provide detailed, insightful analysis.")
                           .prompt("Please analyze the following:\n\n" + text)
@@ -214,7 +214,7 @@ public static void main(String[] args) throws Exception {
               ModelResponse response =
                   genkit.generateStream(
                       GenerateOptions.builder()
-                          .model("xai/grok-4")
+                          .model("xai/grok-4.3")
                           .system(
                               "You are a creative writer with a unique perspective. Write engaging, imaginative content.")
                           .prompt(prompt)

From f3be19cf737595df6b4afc45631af1e16da1d7ba Mon Sep 17 00:00:00 2001
From: xavidop 
Date: Mon, 6 Jul 2026 19:36:40 +0200
Subject: [PATCH 5/6] feat: mistral update

---
 plugins/mistral/README.md                                       | 2 +-
 .../java/com/google/genkit/plugins/mistral/MistralPlugin.java   | 2 --
 .../com/google/genkit/plugins/mistral/MistralPluginTest.java    | 2 +-
 3 files changed, 2 insertions(+), 4 deletions(-)

diff --git a/plugins/mistral/README.md b/plugins/mistral/README.md
index 6f06009ff..7d33fbed7 100644
--- a/plugins/mistral/README.md
+++ b/plugins/mistral/README.md
@@ -16,7 +16,7 @@ This plugin provides integration with Mistral AI models.
 - `mistral-large-2512` - Flagship model (256K context)
 - `mistral-medium-2604` - Mistral Medium 3.5 (128K context)
 - `mistral-small-2603` - Mistral Small 4 (128K context)
-- `magistral-medium-2509`, `magistral-small-2509` - Reasoning models
+- `magistral-medium-2509` - Reasoning model
 - `ministral-3b-2512`, `ministral-8b-2512`, `ministral-14b-2512` - Compact models
 - `codestral-2508` - Code generation specialist (256K context)
 - `devstral-2512` - Developer/agentic coding model
diff --git a/plugins/mistral/src/main/java/com/google/genkit/plugins/mistral/MistralPlugin.java b/plugins/mistral/src/main/java/com/google/genkit/plugins/mistral/MistralPlugin.java
index 6d4ecd838..2c8990196 100644
--- a/plugins/mistral/src/main/java/com/google/genkit/plugins/mistral/MistralPlugin.java
+++ b/plugins/mistral/src/main/java/com/google/genkit/plugins/mistral/MistralPlugin.java
@@ -46,11 +46,9 @@ public class MistralPlugin implements Plugin {
           "mistral-medium-2604",
           "mistral-medium-2508",
           "mistral-small-2603",
-          "mistral-small-2506",
 
           // Reasoning models
           "magistral-medium-2509",
-          "magistral-small-2509",
 
           // Compact models
           "ministral-3b-2512",
diff --git a/plugins/mistral/src/test/java/com/google/genkit/plugins/mistral/MistralPluginTest.java b/plugins/mistral/src/test/java/com/google/genkit/plugins/mistral/MistralPluginTest.java
index d7feab971..63d9303f3 100644
--- a/plugins/mistral/src/test/java/com/google/genkit/plugins/mistral/MistralPluginTest.java
+++ b/plugins/mistral/src/test/java/com/google/genkit/plugins/mistral/MistralPluginTest.java
@@ -89,7 +89,7 @@ void testInitializesActions() {
   void testSupportedModels() {
     assertNotNull(MistralPlugin.SUPPORTED_MODELS);
     assertTrue(MistralPlugin.SUPPORTED_MODELS.contains("mistral-large-2512"));
-    assertTrue(MistralPlugin.SUPPORTED_MODELS.contains("mistral-small-2506"));
+    assertTrue(MistralPlugin.SUPPORTED_MODELS.contains("mistral-small-2603"));
   }
 
   @Test

From a382d65d76ccd641759565cf212435fc40f58b0a Mon Sep 17 00:00:00 2001
From: xavidop 
Date: Mon, 6 Jul 2026 19:50:18 +0200
Subject: [PATCH 6/6] fix: gemini feedback

---
 .../plugins/awsbedrock/AwsBedrockEmbedder.java   |  9 +++------
 .../genkit/plugins/chroma/ChromaVectorStore.java | 16 ++++++++++++++--
 .../plugins/compatoai/CompatOAIEmbedder.java     | 11 +++--------
 .../genkit/plugins/milvus/MilvusVectorStore.java | 16 ++++++++++++++--
 .../genkit/plugins/mongodb/MongoVectorStore.java | 15 +++++++++++++--
 .../mongodb/session/MongoSessionStore.java       |  2 +-
 .../genkit/plugins/qdrant/QdrantVectorStore.java | 16 ++++++++++++++--
 7 files changed, 62 insertions(+), 23 deletions(-)

diff --git a/plugins/aws-bedrock/src/main/java/com/google/genkit/plugins/awsbedrock/AwsBedrockEmbedder.java b/plugins/aws-bedrock/src/main/java/com/google/genkit/plugins/awsbedrock/AwsBedrockEmbedder.java
index 4b884f637..b6faf3fd2 100644
--- a/plugins/aws-bedrock/src/main/java/com/google/genkit/plugins/awsbedrock/AwsBedrockEmbedder.java
+++ b/plugins/aws-bedrock/src/main/java/com/google/genkit/plugins/awsbedrock/AwsBedrockEmbedder.java
@@ -35,8 +35,6 @@
 import okhttp3.OkHttpClient;
 import okhttp3.Request;
 import okhttp3.Response;
-import org.slf4j.Logger;
-import org.slf4j.LoggerFactory;
 
 /**
  * AWS Bedrock embedder implementation for Genkit.
@@ -47,8 +45,6 @@
  */
 public class AwsBedrockEmbedder extends Embedder {
 
-  private static final Logger logger = LoggerFactory.getLogger(AwsBedrockEmbedder.class);
-
   private final String modelId;
   private final AwsBedrockPluginOptions options;
   private final OkHttpClient client;
@@ -109,8 +105,9 @@ public EmbedResponse run(ActionContext context, EmbedRequest request) {
       for (Document doc : request.getDocuments()) {
         String text = doc.text();
         if (text == null || text.isEmpty()) {
-          logger.warn("Document has empty text, skipping");
-          continue;
+          // Throw rather than skip: skipping would make the embeddings list shorter than the
+          // documents list, breaking the 1-to-1 index mapping the caller relies on.
+          throw new GenkitException("Document text cannot be null or empty");
         }
         embeddings.add(embedOne(text));
       }
diff --git a/plugins/chroma/src/main/java/com/google/genkit/plugins/chroma/ChromaVectorStore.java b/plugins/chroma/src/main/java/com/google/genkit/plugins/chroma/ChromaVectorStore.java
index ea1700cdd..daef61438 100644
--- a/plugins/chroma/src/main/java/com/google/genkit/plugins/chroma/ChromaVectorStore.java
+++ b/plugins/chroma/src/main/java/com/google/genkit/plugins/chroma/ChromaVectorStore.java
@@ -200,15 +200,27 @@ public IndexerResponse index(ActionContext context, IndexerRequest request) {
     }
     String id = ensureCollection();
 
+    // Batch-generate embeddings for all documents in a single embedder call.
+    EmbedResponse embedResponse = embedder.run(context, new EmbedRequest(documents));
+    if (embedResponse.getEmbeddings() == null
+        || embedResponse.getEmbeddings().size() != documents.size()) {
+      throw new RuntimeException("Failed to generate embeddings: mismatched output size");
+    }
+
     ObjectNode body = MAPPER.createObjectNode();
     ArrayNode ids = body.putArray("ids");
     ArrayNode embeddings = body.putArray("embeddings");
     ArrayNode contents = body.putArray("documents");
     ArrayNode metadatas = body.putArray("metadatas");
 
-    for (Document doc : documents) {
+    for (int i = 0; i < documents.size(); i++) {
+      Document doc = documents.get(i);
       String content = doc.text() != null ? doc.text() : "";
-      List embedding = generateEmbedding(context, content);
+      float[] values = embedResponse.getEmbeddings().get(i).getValues();
+      List embedding = new ArrayList<>(values.length);
+      for (float v : values) {
+        embedding.add(v);
+      }
       ids.add(getOrGenerateId(doc));
       embeddings.add(floatsToArray(embedding));
       contents.add(content);
diff --git a/plugins/compat-oai/src/main/java/com/google/genkit/plugins/compatoai/CompatOAIEmbedder.java b/plugins/compat-oai/src/main/java/com/google/genkit/plugins/compatoai/CompatOAIEmbedder.java
index 5c2336f4a..79b76581f 100644
--- a/plugins/compat-oai/src/main/java/com/google/genkit/plugins/compatoai/CompatOAIEmbedder.java
+++ b/plugins/compat-oai/src/main/java/com/google/genkit/plugins/compatoai/CompatOAIEmbedder.java
@@ -34,8 +34,6 @@
 import java.util.List;
 import java.util.concurrent.TimeUnit;
 import okhttp3.*;
-import org.slf4j.Logger;
-import org.slf4j.LoggerFactory;
 
 /**
  * Embedder implementation for any provider that exposes an OpenAI-compatible {@code /embeddings}
@@ -46,7 +44,6 @@
  */
 public class CompatOAIEmbedder extends Embedder {
 
-  private static final Logger logger = LoggerFactory.getLogger(CompatOAIEmbedder.class);
   private static final MediaType JSON_MEDIA_TYPE = MediaType.parse("application/json");
 
   private final String apiModelName;
@@ -128,14 +125,12 @@ private EmbedResponse callApi(EmbedRequest request) throws IOException {
     for (Document doc : request.getDocuments()) {
       String text = doc.text();
       if (text == null || text.isEmpty()) {
-        logger.warn("Document has empty text, skipping");
-        continue;
+        // Throw rather than skip: skipping would make the returned embeddings list shorter than the
+        // documents list, breaking the 1-to-1 index mapping the caller relies on.
+        throw new GenkitException("Document text cannot be null or empty");
       }
       input.add(text);
     }
-    if (input.isEmpty()) {
-      throw new GenkitException("No valid documents to embed - all documents had empty text");
-    }
 
     Request.Builder requestBuilder =
         new Request.Builder()
diff --git a/plugins/milvus/src/main/java/com/google/genkit/plugins/milvus/MilvusVectorStore.java b/plugins/milvus/src/main/java/com/google/genkit/plugins/milvus/MilvusVectorStore.java
index 222cba462..c5c2fc063 100644
--- a/plugins/milvus/src/main/java/com/google/genkit/plugins/milvus/MilvusVectorStore.java
+++ b/plugins/milvus/src/main/java/com/google/genkit/plugins/milvus/MilvusVectorStore.java
@@ -189,12 +189,24 @@ public IndexerResponse index(ActionContext context, IndexerRequest request) {
       return new IndexerResponse();
     }
 
+    // Batch-generate embeddings for all documents in a single embedder call.
+    EmbedResponse embedResponse = embedder.run(context, new EmbedRequest(documents));
+    if (embedResponse.getEmbeddings() == null
+        || embedResponse.getEmbeddings().size() != documents.size()) {
+      throw new RuntimeException("Failed to generate embeddings: mismatched output size");
+    }
+
     ObjectNode body = MAPPER.createObjectNode();
     body.put("collectionName", config.getCollectionName());
     ArrayNode data = body.putArray("data");
-    for (Document doc : documents) {
+    for (int i = 0; i < documents.size(); i++) {
+      Document doc = documents.get(i);
       String content = doc.text() != null ? doc.text() : "";
-      List embedding = generateEmbedding(context, content);
+      float[] values = embedResponse.getEmbeddings().get(i).getValues();
+      List embedding = new ArrayList<>(values.length);
+      for (float v : values) {
+        embedding.add(v);
+      }
 
       ObjectNode entity = data.addObject();
       entity.set(VECTOR_FIELD, floatsToArray(embedding));
diff --git a/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/MongoVectorStore.java b/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/MongoVectorStore.java
index 431db234f..4dc03ed1a 100644
--- a/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/MongoVectorStore.java
+++ b/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/MongoVectorStore.java
@@ -291,10 +291,21 @@ public IndexerResponse index(ActionContext context, IndexerRequest request) {
       logger.warn("No documents to index");
       return new IndexerResponse();
     }
+    // Batch-generate embeddings for all documents in a single embedder call.
+    EmbedResponse embedResponse = embedder.run(context, new EmbedRequest(documents));
+    if (embedResponse.getEmbeddings() == null
+        || embedResponse.getEmbeddings().size() != documents.size()) {
+      throw new RuntimeException("Failed to generate embeddings: mismatched output size");
+    }
     int indexed = 0;
-    for (Document doc : documents) {
+    for (int i = 0; i < documents.size(); i++) {
+      Document doc = documents.get(i);
       String content = doc.text() != null ? doc.text() : "";
-      List embedding = generateEmbedding(context, content);
+      float[] values = embedResponse.getEmbeddings().get(i).getValues();
+      List embedding = new ArrayList<>(values.length);
+      for (float v : values) {
+        embedding.add((double) v);
+      }
       String id = getOrGenerateId(doc);
 
       org.bson.Document stored = new org.bson.Document("_id", id);
diff --git a/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/session/MongoSessionStore.java b/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/session/MongoSessionStore.java
index be788130e..bb36fafd4 100644
--- a/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/session/MongoSessionStore.java
+++ b/plugins/mongodb/src/main/java/com/google/genkit/plugins/mongodb/session/MongoSessionStore.java
@@ -623,7 +623,7 @@ private Row readRow(String prefix, String id) {
     payload.remove("pk");
     payload.remove("version");
     try {
-      return new Row((ObjectNode) MAPPER.readTree(payload.toJson()), version);
+      return new Row((ObjectNode) MAPPER.valueToTree(payload), version);
     } catch (Exception e) {
       throw new GenkitException("Failed to read session document " + id + ": " + e.getMessage(), e);
     }
diff --git a/plugins/qdrant/src/main/java/com/google/genkit/plugins/qdrant/QdrantVectorStore.java b/plugins/qdrant/src/main/java/com/google/genkit/plugins/qdrant/QdrantVectorStore.java
index 6e7dfc73a..e0701c3da 100644
--- a/plugins/qdrant/src/main/java/com/google/genkit/plugins/qdrant/QdrantVectorStore.java
+++ b/plugins/qdrant/src/main/java/com/google/genkit/plugins/qdrant/QdrantVectorStore.java
@@ -187,11 +187,23 @@ public IndexerResponse index(ActionContext context, IndexerRequest request) {
       return new IndexerResponse();
     }
 
+    // Batch-generate embeddings for all documents in a single embedder call.
+    EmbedResponse embedResponse = embedder.run(context, new EmbedRequest(documents));
+    if (embedResponse.getEmbeddings() == null
+        || embedResponse.getEmbeddings().size() != documents.size()) {
+      throw new RuntimeException("Failed to generate embeddings: mismatched output size");
+    }
+
     ObjectNode body = MAPPER.createObjectNode();
     ArrayNode points = body.putArray("points");
-    for (Document doc : documents) {
+    for (int i = 0; i < documents.size(); i++) {
+      Document doc = documents.get(i);
       String content = doc.text() != null ? doc.text() : "";
-      List embedding = generateEmbedding(context, content);
+      float[] values = embedResponse.getEmbeddings().get(i).getValues();
+      List embedding = new ArrayList<>(values.length);
+      for (float v : values) {
+        embedding.add(v);
+      }
 
       ObjectNode point = points.addObject();
       point.put("id", getOrGenerateId(doc));