added llama-server inference support

src/memory/core.py

@@ -107,6 +107,18 @@ def _create_embedding_provider(self) -> EmbeddingProvider:
                 model=self.config.embedding.model,
                 base_url=self.config.embedding.base_url or "http://localhost:11434",
             )
+        elif provider == "llama":
+            from memory.embeddings.llama import LlamaEmbedding
+            return LlamaEmbedding(
+                model=self.config.embedding.model,
+                base_url=self.config.embedding.base_url or "http://localhost:11435",
+            )
+        elif provider == "llama-nomic":
+            from memory.embeddings.llama_nomic import LlamaNomicEmbedding
+            return LlamaNomicEmbedding(
+                model=self.config.embedding.model,
+                base_url=self.config.embedding.base_url or "http://localhost:11435",
+            )
         elif provider == "openai":
             from memory.embeddings.openai_embed import OpenAIEmbedding
             return OpenAIEmbedding(

src/memory/embeddings/__init__.py

@@ -1,9 +1,13 @@
 from memory.embeddings.base import EmbeddingProvider
 from memory.embeddings.ollama import OllamaEmbedding
 from memory.embeddings.openai_embed import OpenAIEmbedding
+from memory.embeddings.llama import LlamaEmbedding
+from memory.embeddings.llama_nomic import LlamaNomicEmbedding
 
 __all__ = [
     "EmbeddingProvider",
     "OllamaEmbedding",
     "OpenAIEmbedding",
+    "LlamaEmbedding",
+    "LlamaNomicEmbedding"
 ]

src/memory/embeddings/base.py

@@ -6,5 +6,9 @@ class EmbeddingProvider(ABC):
     def embed(self, text: str) -> list[float]:
         ...
 
+    @abstractmethod
+    def search(self, text: str) -> list[float]:
+        ...
+
     def embed_batch(self, texts: list[str]) -> list[list[float]]:
         return [self.embed(t) for t in texts]

src/memory/embeddings/llama.py

@@ -0,0 +1,27 @@
+import httpx
+from memory.embeddings.base import EmbeddingProvider
+
+
+class LlamaEmbedding(EmbeddingProvider):
+    def __init__(self, model: str = "text-embedder",
+                 base_url: str = "http://localhost:11435"):
+        self.model = model
+        self.base_url = base_url
+
+    def embed(self, text: str) -> list[float]:
+        resp = httpx.post(
+            f"{self.base_url}/embeddings",
+            json={"model": self.model, "content": text},
+            timeout=30.0,
+        )
+        resp.raise_for_status()
+        return resp.json()[0]["embedding"][0]
+
+    def search(self, text: str) -> list[float]:
+        resp = httpx.post(
+            f"{self.base_url}/embeddings",
+            json={"model": self.model, "content": text},
+            timeout=30.0,
+        )
+        resp.raise_for_status()
+        return resp.json()[0]["embedding"][0]

src/memory/embeddings/llama_nomic.py

@@ -0,0 +1,27 @@
+import httpx
+from memory.embeddings.base import EmbeddingProvider
+
+
+class LlamaNomicEmbedding(EmbeddingProvider):
+    def __init__(self, model: str = "text-embedder",
+                 base_url: str = "http://localhost:11435"):
+        self.model = model
+        self.base_url = base_url
+
+    def embed(self, text: str) -> list[float]:
+        resp = httpx.post(
+            f"{self.base_url}/embeddings",
+            json={"model": self.model, "content": 'search_document: ' + text},
+            timeout=30.0,
+        )
+        resp.raise_for_status()
+        return resp.json()[0]["embedding"][0]
+
+    def search(self, text: str) -> list[float]:
+        resp = httpx.post(
+            f"{self.base_url}/embeddings",
+            json={"model": self.model, "content": 'search_query: ' + text},
+            timeout=30.0,
+        )
+        resp.raise_for_status()
+        return resp.json()[0]["embedding"][0]

src/memory/embeddings/ollama.py

@@ -37,3 +37,5 @@ def embed(self, text: str) -> list[float]:
         )
         resp.raise_for_status()
         return resp.json()["embedding"]
+
+    search = embed

src/memory/embeddings/openai_embed.py

@@ -17,3 +17,5 @@ def embed(self, text: str) -> list[float]:
         )
         resp.raise_for_status()
         return resp.json()["data"][0]["embedding"]
+
+    search = embed

src/memory/search.py

@@ -99,7 +99,7 @@ def tiered_search(
 
     # FTS results are sparse — fall back to hybrid (embed + vector search + merge)
     try:
-        query_vec = embedding_provider.embed(query)
+        query_vec = embedding_provider.search(query)
         vec_results = db.vector_search(
             query_vec, limit=limit * 2, project=project, source=source
         )
@@ -144,7 +144,7 @@ def hybrid_search(
                 r["score"] = r["score"] / max_score if max_score > 0 else 0.0
         return fts_results[:limit]
 
-    query_vec = embedding_provider.embed(query)
+    query_vec = embedding_provider.search(query)
     vec_results = db.vector_search(
         query_vec, limit=limit * 2, project=project, source=source
     )