release: v0.13.0 — error unification, LLM tool forwarding, MCP spec compliance

.github/workflows/ci.yml

@@ -43,8 +43,24 @@ jobs:
           - "--no-default-features --features store"
           - "--no-default-features --features config"
           - "--no-default-features --features client-async"
+          - "--no-default-features --features cache"
+          - "--no-default-features --features discovery-async"
           - "--no-default-features --features graph"
+          - "--no-default-features --features graph-async"
+          - "--no-default-features --features graph-pool"
+          - "--no-default-features --features graph-cjk"
           - "--no-default-features --features graph-pg"
+          - "--no-default-features --features mcp"
+          - "--no-default-features --features mcp-http"
+          - "--no-default-features --features tokens"
+          - "--no-default-features --features safety"
+          - "--no-default-features --features telemetry"
+          - "--no-default-features --features search"
+          - "--no-default-features --features federation"
+          - "--no-default-features --features embedding"
+          - "--no-default-features --features embedding-openai"
+          - "--no-default-features --features vector-index"
+          - "--no-default-features --features install"
           - "--no-default-features --features qdrant"
           - "--no-default-features --features elastic"
           - "--no-default-features --features full"

AGENTS.md

@@ -8,7 +8,7 @@
 
 | Command | Description |
 |---------|-------------|
-| `cargo test --all-features` | Run all tests (483 passed, 12 ignored) |
+| `cargo test --all-features` | Run all tests (599 passed, 13 ignored) |
 | `cargo clippy --all-features -- -D warnings` | Lint |
 | `cargo fmt --all -- --check` | Format check |
 | `cargo bench` | Run criterion benchmarks |

CHANGELOG.md

@@ -5,6 +5,33 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.13.0] - 2026-07-03
+
+### Added
+
+- **llm**: `LLMRequest::tools` and `LLMRequest::response_format` are now **forwarded to the provider APIs**. OpenAI receives `tools` (`type: "function"`) and `response_format` (`json_object` / `json_schema`); Anthropic receives `tools` (with `input_schema`) and, for `ResponseFormat::JsonSchema`, `output_config.format`. Previously both fields were accepted by the builder but silently dropped.
+- **llm**: `LLMResponse::tool_calls: Vec<ToolCall>` — tool calls the model requested are parsed back from OpenAI `tool_calls` and Anthropic `tool_use` content blocks. `LLMResponse` now also captures `finish_reason` (OpenAI `finish_reason` / Anthropic `stop_reason`), `id`, and `created` from the provider response.
+- **mcp**: protocol-version negotiation — `initialize` echoes the client's requested `protocolVersion` when supported (`2025-06-18`, `2025-03-26`, `2024-11-05`), otherwise proposes the server's latest (`2025-06-18`). Exposed via `McpServer::negotiate_protocol_version` and the `SUPPORTED_PROTOCOL_VERSIONS` / `LATEST_PROTOCOL_VERSION` constants.
+- **mcp**: `ping` method (returns `{}`), **prompts** support (`prompts/list`, `prompts/get`, `McpServer::register_prompt` / `set_prompt_handler`, `PromptDescription` / `PromptArgument`), and `resources/templates/list`. The `prompts` capability is advertised in `initialize` when prompts are registered. Both stdio and HTTP/SSE transports support all new methods.
+- **error**: `KernelError::Embedding` and `KernelError::Discovery` variants (with `KernelError::embedding` / `KernelError::discovery` constructors).
+
+### Changed
+
+- **error** (**breaking**): the `embedding` and `discovery` subsystems now return `crate::error::Result` (`KernelError`) instead of `anyhow::Result` — the `EmbeddingProvider` / `VectorIndex` / `AsyncVectorIndex` traits, all provider and index constructors (`FastembedProvider`, `OpenAIEmbeddingClient`, `Qwen3Provider`, `NomicMoeProvider`, `LazyFastembedProvider`, `TurbovecIndex`, `QdrantVectorIndex`, `ElasticsearchVectorIndex`), `DiscoverySource`, `chunk_batch`, the `discovery::fetch*` functions, and `provider::sync::*`. `anyhow` no longer appears in the library's public surface. Downstream code that matched on `anyhow::Error` must switch to `KernelError`.
+- **mcp** (**breaking**): `McpServer::initialize_response` now takes the client's requested protocol version (`initialize_response(Option<&str>)`).
+- **mcp**: `ToolDescription` and `ResourceDescription` now serialize with the correct MCP wire-format field names — `inputSchema` (was `input_schema`) and `mimeType` (was `mime_type`).
+- **mcp**: JSON-RPC request `id`s are preserved verbatim (string **or** number) in responses, per JSON-RPC 2.0 — previously only integer ids round-tripped.
+- **mcp**: `tools/call` reports **tool-execution failures in-band** as a result with `isError: true` (so the model can react), and reserves the JSON-RPC error path (`-32602`) for an unknown tool — matching the MCP spec.
+
+### Fixed
+
+- **embedding**: `LazyFastembedProvider::embed_batch` no longer panics with an index-out-of-bounds when the inner provider returns fewer vectors than inputs (a truncated/malformed response); it now returns a `KernelError::Embedding`.
+- **llm**: `CacheClient::complete` offloads the synchronous `KvStore` read/write to `tokio::task::spawn_blocking`, so a slow or remote store (or a single-threaded runtime) no longer blocks the async reactor on the completion hot path.
+
+### CI
+
+- Isolated per-feature build/test matrix entries added for `cache`, `discovery-async`, `graph-async`, `graph-pool`, `graph-cjk`, `mcp`, `mcp-http`, `tokens`, `safety`, `telemetry`, `search`, `federation`, `embedding`, `embedding-openai`, `vector-index`, and `install`, so a missing `#[cfg]` gate is caught even when a sibling feature isn't co-enabled.
+
 ## [0.12.0] - 2026-07-02
 
 ### Changed
@@ -27,6 +54,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 - **graph**: new optional `graph-pg-tls` feature adding TLS support to `PgGraph` connections, closing #48. `PgGraph::connect_native_tls(url)` is a one-call convenience constructor using `native-tls` with the system trust store (full certificate chain and hostname verification, not weakened) — covers the common case of a Postgres server requiring `sslmode=require`+ (e.g. RDS with `rds.force_ssl`). `PgGraph::connect_tls` / `connect_config_tls` are generic over any `postgres::tls::MakeTlsConnect` implementor for custom CAs, client certificates, or a caller-vendored connector. Existing `connect` / `connect_config` (`NoTls`) are unchanged — fully backward compatible, no new mandatory deps for `graph-pg` consumers.
 
+
 ## [0.10.0] - 2026-06-29
 
 ### Added

CODE_OF_CONDUCT.md

@@ -58,8 +58,10 @@ representative at an online or offline event.
 ## Enforcement
 
 Instances of abusive, harassing, or otherwise unacceptable behavior may be
-reported to the community leaders responsible for enforcement at
-[INSERT CONTACT METHOD].
+reported to the project maintainers via GitHub — open a private report through
+the repository's **Security** tab ("Report a vulnerability") or contact a
+maintainer of the [`epicsagas/llm-kernel`](https://github.com/epicsagas/llm-kernel)
+repository directly.
 All complaints will be reviewed and investigated promptly and fairly.
 
 All community leaders are obligated to respect the privacy and security of the

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name         = "llm-kernel"
-version      = "0.12.0"
+version      = "0.13.0"
 edition      = "2024"
 description  = "Foundation library for Rust AI-native apps — provider catalog, LLM client, MCP server, search, telemetry, and safety"
 license      = "Apache-2.0"
@@ -152,9 +152,12 @@ safety = ["dep:regex"]
 eval = ["dep:clap", "tokens", "safety", "embedding", "search"]
 eval-full = ["eval", "graph"]
 
-# Everything except Windows-only execution backends.
-# embedding-fastembed-directml is intentionally excluded: it pulls in the ort
-# RC dependency which is not appropriate for cross-platform builds.
+# Everything except Windows-only execution backends and dev-only CLI tooling.
+# Intentionally excluded:
+#   - embedding-fastembed-directml: pulls in the ort RC dependency, Windows-only.
+#   - embedding-fastembed-dynamic-linking: opt-in dynamic ONNX linking (see #50).
+#   - eval / eval-full: developer CLI tooling (the `llm-kernel-eval` binary),
+#     not part of the library surface. Enable explicitly to run the eval suite.
 full = ["provider", "discovery", "discovery-async", "client-async", "cache", "secrets", "store", "config", "graph", "graph-async", "graph-pool", "graph-cjk", "graph-pg", "graph-pg-tls", "mcp", "mcp-http", "tokens", "install", "search", "embedding", "embedding-openai", "embedding-fastembed", "embedding-fastembed-qwen3", "embedding-fastembed-nomic-moe", "vector-index", "qdrant", "elastic", "federation", "telemetry", "safety", "catalog-sync"]
 
 [dependencies]

PROGRESS.md

@@ -1,24 +1,42 @@
 # Progress
 
-> Auto-generated status snapshot. Last updated: 2026-06-15
+> Auto-generated status snapshot. Last updated: 2026-07-03
 
-## Current Version: v0.9.0
+## Current Version: v0.13.0
 
 | Metric | Value |
 |--------|-------|
-| Version | `0.9.0` |
+| Version | `0.13.0` |
 | Edition | Rust 2024, MSRV 1.92 |
-| Lines of code | ~18,100 |
-| Total tests | 511 (499 passed, 12 ignored, 0 failed) |
+| Lines of code | ~23,000 |
+| Total tests | `--all-features`: 599 passed, 13 ignored, 0 failed |
 | Backend features | `graph-pg` (PostgreSQL), `qdrant` (vector search), `elastic` (Elasticsearch vector search) |
-| Open PRs | 1 |
-| Open branches | 1 |
-| Last commit | `feat: v0.9.0 search integrations` |
+| Last commit | `chore: bump v0.13.0` |
 
 ---
 
 ## Recent Releases
 
+### v0.13.0 (2026-07-03)
+
+- **error**: unified `embedding` + `discovery` public APIs onto `KernelError` (new `Embedding` / `Discovery` variants) — no more `anyhow::Result` in the library's public surface (**breaking**)
+- **llm**: `LLMRequest::tools` and `response_format` are now forwarded to OpenAI and Anthropic; tool calls are parsed back into `LLMResponse::tool_calls`
+- **mcp**: protocol version negotiation (2025-06-18), `ping`, prompts (`prompts/list` / `prompts/get`), string/number JSON-RPC ids, `tools/call` in-band `isError`, and camelCase wire format (`inputSchema` / `mimeType`)
+- **embedding**: fixed a `LazyFastembedProvider::embed_batch` panic on a truncated provider response; `CacheClient` now offloads blocking store I/O via `spawn_blocking`
+- **ci**: isolated per-feature build/test matrix entries (`mcp`, `tokens`, `safety`, `search`, `cache`, …)
+
+### v0.12.0 (2026-07-02)
+
+- **embedding** (breaking): `ModelState::Failed(String)` → `Failed { message, panicked }`; dropped default `ort-load-dynamic` so `embedding-fastembed` statically links ONNX Runtime, made the model-load path panic-safe via `catch_unwind`, and added `LazyFastembedProvider::reset()` + opt-in `embedding-fastembed-dynamic-linking` (#50)
+
+### v0.11.0 (2026-07-01)
+
+- **graph** (`graph-pg-tls`): TLS support for `PgGraph` connections — `connect_native_tls` / `connect_tls` / `connect_config_tls` (#48)
+
+### v0.10.0 (2026-06-29)
+
+- **graph**: pure-Rust CSR graph algorithms (`algo/`) — weighted PageRank, connected components, label propagation, Dijkstra, Jaccard/Adamic-Adar similarity; `smart_recall`'s graph boost now ranks by true PageRank centrality (SQLite + PostgreSQL share one impl)
+
 ### v0.9.0 (2026-06-15)
 
 - **embedding** (`elastic`): `ElasticsearchVectorIndex` — `AsyncVectorIndex` over Elasticsearch 8.x (dense_vector cosine, bulk upsert/delete, knn `_search`, `_count`). 공식 `elasticsearch` 크레이트가 alpha-only라 **직접 구현한 reqwest 클라이언트** 사용 (v1.0.0 semver lock 안전)
@@ -151,8 +169,8 @@ safety       → secret masking, error classification, prompt-injection detectio
 
 | Check | Status |
 |-------|--------|
-| All tests pass | ✅ 489/489 (477 passed, 12 ignored) |
+| All tests pass | ✅ 593 passed, 13 ignored, 0 failed (`--all-features`) |
 | Clippy clean | ✅ (verified before each release) |
 | CI passing | ✅ Linux + macOS dual runner |
 | Crate structure | ✅ Monolithic (subcrate removed) |
-| Roadmap on track | ✅ v0.7.0 complete, v0.8.0 ready to start |
+| Roadmap on track | ✅ v0.13.0 complete, v1.0.0 next |

QUICKSTART.md

@@ -13,21 +13,21 @@ Add to your `Cargo.toml`. The `provider` feature is enabled by default.
 
 ```toml
 [dependencies]
-llm-kernel = "0.6.0"
+llm-kernel = "0.13.0"
 ```
 
 For the async LLM client:
 
 ```toml
 [dependencies]
-llm-kernel = { version = "0.6.0", features = ["client-async"] }
+llm-kernel = { version = "0.13.0", features = ["client-async"] }
 ```
 
 To enable everything:
 
 ```toml
 [dependencies]
-llm-kernel = { version = "0.6.0", features = ["full"] }
+llm-kernel = { version = "0.13.0", features = ["full"] }
 ```
 
 ## 2. Browse the provider catalog
@@ -82,7 +82,7 @@ println!("{}", response.content);
 
 ```toml
 [dependencies]
-llm-kernel = { version = "0.6.0", features = ["graph"] }
+llm-kernel = { version = "0.13.0", features = ["graph"] }
 ```
 
 ```rust
@@ -117,7 +117,7 @@ for scored in &results {
 
 ```toml
 [dependencies]
-llm-kernel = { version = "0.6.0", features = ["embedding-fastembed"] }
+llm-kernel = { version = "0.13.0", features = ["embedding-fastembed"] }
 ```
 
 ```rust

README.md

@@ -23,8 +23,8 @@ llm-kernel provides the foundational layer for building LLM-powered tools, agent
 - **Model discovery** — dynamic model discovery from models.dev, Ollama, OpenAI-compatible endpoints
 - **Credential vault** — dotenv-style API key management with atomic writes
 - **Config loader** — TOML config with auto-create from template
-- **Knowledge graph** — `GraphBackend` trait (SQLite impl), FTS5 search, smart recall, BFS traversal, CJK search, schema migrations, async wrappers
-- **MCP server** — JSON-RPC 2.0 server framework with stdio and HTTP/SSE transports, async handlers, Bearer auth
+- **Knowledge graph** — `GraphBackend` trait (SQLite impl), FTS5 search, smart recall, BFS traversal, CJK search, schema migrations, async wrappers, pure-Rust graph algorithms (PageRank, community detection, shortest path, similarity)
+- **MCP server** — JSON-RPC 2.0 server framework (protocol 2025-06-18) with stdio and HTTP/SSE transports, tools, resources, prompts, `ping`, async handlers, Bearer auth
 - **Key-value store** — `KvStore` trait powering LLM response caching and other byte-oriented stores
 - **Embedding** — provider trait + cosine similarity, local ONNX (44 models), Qwen3 candle, Nomic V2 MoE candle, OpenAI remote, compressed vector indexing ([full model list →](EMBEDDING_MODELS.md))
 - **Search** — Reciprocal Rank Fusion for hybrid search result merging
@@ -46,12 +46,12 @@ Each module is gated behind a feature flag so you only pay for what you use.
 | `secrets` | SecretVault credential management | |
 | `store` | SQLite init helpers (WAL, FTS5, schema versioning) + `KvStore` | |
 | `config` | TOML config loader | |
-| `graph` | Knowledge graph — `GraphBackend` trait, SQLite impl, FTS5, smart recall, BFS, migrations | |
+| `graph` | Knowledge graph — `GraphBackend` trait, SQLite impl, FTS5, smart recall, BFS, migrations, graph algorithms (PageRank, community, shortest-path, similarity) | |
 | `graph-async` | Async graph wrappers (requires tokio) | |
 | `graph-pool` | Multi-connection async graph pool (`AsyncPoolGraph`, WAL concurrency) | |
 | `graph-cjk` | CJK-aware graph search via Rust-side segmentation (no schema change) | |
 | `graph-pg` | PostgreSQL `GraphBackend` (`PgGraph`) + SQLite↔PostgreSQL migration CLI | |
-| `mcp` | MCP server — JSON-RPC 2.0, stdio transport, async handlers, Bearer auth | |
+| `mcp` | MCP server — JSON-RPC 2.0 (protocol 2025-06-18), stdio transport, tools/resources/prompts, `ping`, async handlers, Bearer auth | |
 | `mcp-http` | MCP remote transport — HTTP/SSE (axum + tokio) | |
 | `cache` | LLM response cache — `CacheClient` over `KvStore` | |
 | `tokens` | Token estimation, budgeting, and sentence-aware document chunking | |
@@ -80,28 +80,28 @@ Add to your `Cargo.toml`:
 
 ```toml
 [dependencies]
-llm-kernel = "0.9.1"
+llm-kernel = "0.13.0"
 ```
 
 The `provider` feature is enabled by default. For the async client:
 
 ```toml
 [dependencies]
-llm-kernel = { version = "0.9.1", features = ["client-async"] }
+llm-kernel = { version = "0.13.0", features = ["client-async"] }
 ```
 
 For the knowledge graph with async wrappers:
 
 ```toml
 [dependencies]
-llm-kernel = { version = "0.9.1", features = ["graph", "graph-async"] }
+llm-kernel = { version = "0.13.0", features = ["graph", "graph-async"] }
 ```
 
 For local embedding (ONNX, no API key):
 
 ```toml
 [dependencies]
-llm-kernel = { version = "0.9.1", features = ["embedding-fastembed"] }
+llm-kernel = { version = "0.13.0", features = ["embedding-fastembed"] }
 ```
 
 ## Usage
@@ -337,11 +337,11 @@ println!("{} nodes, {} edges", stats.total_nodes, stats.total_edges);
 ### MCP server
 
 ```rust
-use llm_kernel::mcp::{McpServer, Tool, JsonRpcRequest};
+use llm_kernel::mcp::{McpServer, ToolDescription};
 use serde_json::json;
 
 let mut server = McpServer::new("my-server", "1.0.0");
-server.register_tool(Tool {
+server.register_tool(ToolDescription {
     name: "greet".into(),
     description: "Say hello".into(),
     input_schema: json!({
@@ -571,9 +571,9 @@ llm-kernel is a **lightweight foundation layer** — compose it with rig or lang
 - **`EmbeddingProvider` trait** — unified interface for `FastembedProvider` (ONNX), `Qwen3Provider` (candle), `NomicMoeProvider` (candle), `OpenAIEmbeddingClient` (remote)
 - **`VectorIndex` trait** — unified interface for compressed vector indexes; `TurbovecIndex` (TurboQuant) implements 2-bit/4-bit quantized ANN search with SIMD kernels
 - **`ProviderIndex`** — zero-copy access to embedded catalog, queryable by provider or model
-- **`McpServer`** — JSON-RPC 2.0 server with stdio transport, Bearer auth, tool registration
+- **`McpServer`** — JSON-RPC 2.0 server (protocol 2025-06-18) with stdio transport, Bearer auth, tools/resources/prompts registration, `ping`
 - **`SecretVault`** — `HashMap<String, String>` with dotenv load/save and symlink guards
-- **`graph`** — SQLite knowledge graph with FTS5 search, composite scoring recall, BFS traversal, importance decay
+- **`graph`** — SQLite knowledge graph with FTS5 search, composite scoring recall, BFS traversal, importance decay, and pure-Rust CSR graph algorithms (PageRank, connected components, label propagation, Dijkstra, Jaccard/Adamic-Adar similarity)
 - **`TelemetryEvent`** — enum-gated variants for structured observability (no PII)
 - **`safety`** — secret masking, error classification, bidi/ANSI/null sanitization, prompt-injection detection
 - **`SearchProvider`** — unified sync interface for ranking backends; `KeywordIndex` reference impl plus RRF / weighted-sum / CombMNZ fusion