Skip to content

Releases

llm-abi 0.5.2

Choose a tag to compare

@github-actions github-actions released this 28 Aug 16:34
0c14cfd

Patch Changes

  • #44 80577f6 Thanks @jammaru! - Rewrite the published package intro to a short library README: what it is, what you get back, and one compile() example.

llm-abi 0.5.1

Choose a tag to compare

@github-actions github-actions released this 28 Aug 16:18
3808fd0

Patch Changes

  • #42 44a4e88 Thanks @jammaru! - Rewrite the published package README so npm shows the four compatibility labels, the docs site, and that the GPT-5.6 Chat Completions tools example is unsupported. Pin the GitHub Action example to @v0.5.0 instead of @v0.2.0 or @latest.

llm-abi 0.5.0

Choose a tag to compare

@github-actions github-actions released this 28 Aug 15:27
6c26032

Minor Changes

  • #39 8341e88 Thanks @jammaru! - Add experimental vLLM and SGLang runtime schema targets from documented OpenAI-compatible structured outputs. Keyword tables stay example-proven. Default check() and the qwen alias do not change.

llm-abi 0.4.0

Choose a tag to compare

@github-actions github-actions released this 28 Aug 15:11
e0e0a0c

Minor Changes

  • #36 51206c2 Thanks @jammaru! - Add local matrix, lock/diff, check, and a keyword probe suite so loaded runtimes can be compared and snapshotted without secrets, paths, or routing.

llm-abi 0.3.0

Choose a tag to compare

@github-actions github-actions released this 28 Aug 14:39
b1e8694

Minor Changes

  • 153a635 Thanks @jammaru! - Expose target maturity, evidence source, last-verification date, and nightly-adapter coverage on resolved targets, including resolveTarget(). Deduplicate repeated diagnostics and loss items produced by shared IR nodes. Expand npm keywords for additional providers and schema compatibility.

  • 2177b44 Thanks @jammaru! - Add checkRequest() so GPT-5.6 Chat Completions function-tool calls can be rejected before omitted reasoning_effort defaults to medium and the API returns 400.

  • #31 5f7ce66 Thanks @jammaru! - Add a Runtime ABI next to compile() and checkRequest(): checkDeployment(), runtime-scoped schema targets for llama.cpp / LM Studio / Ollama, and a Node-only llm-abi/local doctor and smoke probe. Default check() stays provider-only. The qwen alias remains Alibaba Model Studio.

llm-abi 0.2.0

Choose a tag to compare

@github-actions github-actions released this 28 Aug 05:46
2b7695b

Minor Changes

  • #30 e3ab4f5 Thanks @jammaru! - Add a documented DeepSeek strict-tools target, including $def emit and required-all objects.

  • #35 78aa4c8 Thanks @jammaru! - Add an MCP tools host-subset target that inlines $ref, requires an object root, and keeps stripped constraints in validate.

  • #33 3a66dcb Thanks @jammaru! - Add an sdk-observed Mistral structured target that closes objects without forcing optional fields required.

  • #34 5ec0581 Thanks @jammaru! - Add an OpenRouter gateway target that never claims lossless enforcement for every routed model.

  • #32 bf13078 Thanks @jammaru! - Add a documented Qwen tools target that preserves optional fields and additionalProperties instead of copying OpenAI strict mode.

  • #22 9552e9a Thanks @jammaru! - Report emitted schema size and a conservative token hint from analyze, compile, and check. Unused root $defs now emit a diagnostic instead of disappearing silently.

  • #23 aa9c0f0 Thanks @jammaru! - Accept a closed TypeScript type subset as compile / check / analyze input so the schema ABI is TypeScript-to-provider, not JSON-Schema-only.

  • #31 effa6cb Thanks @jammaru! - Add a documented xAI Grok structured target that omits additionalProperties: false and rejects recursive $ref.

Patch Changes

  • #36 a9e6429 Thanks @jammaru! - Emit enum for TypeScript string, number, and boolean literal unions so MCP and other combinator-hostile hosts receive the object subset.

llm-abi 0.1.0

Choose a tag to compare

@jammaru jammaru released this 28 Aug 05:48
ec2f6cf

Restored release tag after repository recreation.