Skip to content
Permalink

Comparing changes

Choose two branches to see what’s changed or to start a new pull request. If you need to, you can also or learn more about diff comparisons.

Open a pull request

Create a new pull request by comparing changes across two branches. If you need to, you can also . Learn more about diff comparisons here.
base repository: epicsagas/llm-kernel
Failed to load repositories. Confirm that selected base ref is valid, then try again.
Loading
base: v0.12.0
Choose a base ref
...
head repository: epicsagas/llm-kernel
Failed to load repositories. Confirm that selected head ref is valid, then try again.
Loading
compare: v0.13.0
Choose a head ref
  • 1 commit
  • 50 files changed
  • 1 contributor

Commits on Jul 2, 2026

  1. release: v0.13.0 — error unification, LLM tool forwarding, MCP spec c… 

    …ompliance (#53)

* refactor(error)!: return KernelError from embedding and discovery APIs

Unify the public error surface: `embedding` and `discovery` now return
`crate::error::Result` (`KernelError`) instead of `anyhow::Result`, so `anyhow`
no longer leaks through the library's public API. Adds `KernelError::Embedding`
and `KernelError::Discovery` (plus `::embedding` / `::discovery` constructors).

Touches the `EmbeddingProvider` / `VectorIndex` / `AsyncVectorIndex` traits, all
provider and index constructors, the `LazyFastembedProvider` loader path
(`LoadFn` / `load_catching_panics` now yield `KernelError`), `DiscoverySource`,
`chunk_batch`, `discovery::fetch*`, and `provider::sync::*`.

Also guards `LazyFastembedProvider::embed_batch` against an index-out-of-bounds
panic when the inner provider returns fewer vectors than inputs.

BREAKING CHANGE: embedding and discovery APIs return `KernelError` instead of
`anyhow::Error`.

* feat(llm): forward tools and response_format to providers

`LLMRequest::tools` and `response_format` are now sent to the provider APIs
instead of being silently dropped: OpenAI receives `tools` (`type: "function"`)
and `response_format`; Anthropic receives `tools` (with `input_schema`) and, for
`ResponseFormat::JsonSchema`, `output_config.format`. Tool calls are parsed back
into the new `LLMResponse::tool_calls`; `finish_reason`, `id`, and `created` are
captured too.

Also offloads the synchronous `KvStore` read/write in `CacheClient::complete` to
`tokio::task::spawn_blocking`, so a slow or remote store no longer blocks the
async reactor on the hot path.

* feat(mcp): align server with the 2025-06-18 MCP spec

Protocol-version negotiation in `initialize`; `ping`; prompts (`prompts/list`,
`prompts/get`, `register_prompt` / `set_prompt_handler`, `PromptDescription` /
`PromptArgument`); `resources/templates/list`; correct wire-format field names
(`inputSchema` / `mimeType`); verbatim string/number JSON-RPC ids; and in-band
`tools/call` `isError` reporting (reserving `-32602` for an unknown tool).
Applies to both the stdio and HTTP/SSE transports.

* chore(release): v0.13.0

- docs: bump version pins to 0.13.0 (README, QUICKSTART, 10 i18n READMEs); fix
  the MCP quick-start example (`Tool`/`JsonRpcRequest` → `ToolDescription`)
  across all 11 files; propagate the v0.10.0 graph-algorithm content into every
  feature table; refresh ROADMAP/PROGRESS (incl. the real v0.11.0/v0.12.0
  releases); fix the SECURITY supported-versions table and the CODE_OF_CONDUCT
  enforcement contact
- ci: isolated per-feature build/test matrix entries so a missing `#[cfg]` gate
  is caught even when a sibling feature is not co-enabled
- Cargo.toml: document the eval/eval-full and dynamic-linking exclusions from `full`
- bump version to 0.13.0, refresh CHANGELOG and lockfile
    @epicsagas
    epicsagas authored Jul 2, 2026
    Configuration menu
    Copy the full SHA
    0d77833 View commit details
    Browse the repository at this point in the history
Loading