Skip to content

Releases: nghyane/llm-mux

v2.0.18

31 Dec 20:42

Choose a tag to compare

Changelog

  • 9b3aa41 chore(make): add docs target for gh-pages deployment
  • 09c1894 docs(agents): add agent orchestration pattern documentation
  • f3c4754 feat(antigravity): add whitelist-based payload validation for Vertex AI
  • 8bceb95 feat(ir): add content filter, logprobs, and stream state types
  • 3744675 feat(routing): add configurable provider priority and model fallbacks
  • c4d0c6b feat(telemetry): add telemetry stub package
  • 620b72f feat(translator): add FromIR registry pattern with adapter wrappers
  • cf47260 feat(translator): add preprocess package for IR transformations
  • 76b5e51 feat(util): count additional IR fields in token estimation
  • 44682a4 fix(antigravity): consolidate JSON schema cleanup for Claude models
  • 5572715 fix(logging): actually set debug log level when debug=true
  • 7aff982 fix(retry): correct FallbackCodes to allow 500 retry before fallback
  • ccd6c2a fix(retry): implement fail-fast for streaming hang prevention
  • b4b4e9a fix(translator): validate tool_use/tool_result pairs and coalesce messages
  • 72b633b refactor(claude): simplify user ID generation
  • 6def212 refactor(executor): use vertex-envelope translator and extract helpers
  • eeaf6c4 refactor(logging): migrate from logrus to slog
  • e9aed5b refactor(provider): improve retry logic with polling-based wait
  • 36d7c98 refactor(translator): add VertexEnvelopeProvider for Antigravity Claude models
  • 71fcf28 refactor(translator): unify RoleTool handling across from_ir converters

v2.0.17

30 Dec 10:12

Choose a tag to compare

Changelog

  • 6f14ea5 chore: add make help and make release
  • 44c2730 chore: ignore dist build artifacts
  • 48bf2bc chore: organize docker scripts and add edge tag support
  • 8a0309d chore: remove build artifact, update gitignore
  • 3e97146 chore: remove unused examples, auths, embedded packages
  • e09ef73 feat: add Makefile and release script for local releases
  • e1e6b90 feat: support make release [cmd] with arguments
  • f8ea028 fix: remove unreliable docker login check
  • 20afbd4 refactor: consolidate sdk packages into internal
  • 261ceeb refactor: remove dead expandPath function, use util.ResolveAuthDir
  • c82b796 refactor: simplify Makefile to build/test/clean only

v2.0.16

29 Dec 15:41

Choose a tag to compare

Changelog

  • fbf91ec fix(streaming): restore estimated input tokens for Claude format via event-driven architecture

v2.0.15

29 Dec 14:56

Choose a tag to compare

Changelog

  • b44a917 fix(gemini): generate tool call ID when missing and handle streaming partialArgs

v2.0.14

29 Dec 09:57

Choose a tag to compare

Changelog

  • fbfa9ad fix(executor): improve streaming stability and resource handling
  • fde5753 fix(translator): handle all numeric types for maxOutputTokens
  • b487f86 refactor(executor): clean up unnecessary and redundant comments
  • 008aef0 refactor(non-streaming): unify response translation with ResponseTranslator
  • 99b09c9 refactor(streaming): migrate OpenAIStreamProcessor to use StreamTranslator
  • fc4ec14 refactor: consolidate GeminiCLI stream processors
  • ac76070 refactor: remove dead TranslateTokenCount function and call sites

v2.0.13

29 Dec 06:59

Choose a tag to compare

Refactor: Unified Stream Translation

Major cleanup of streaming infrastructure with modular components.

New Components

  • StreamContext - Single source of truth for streaming state (replaces 3 separate state structs)
  • ChunkBufferStrategy - Pluggable buffering (Passthrough/DelayOne) for format-specific needs
  • StreamTranslator - Unified IR-to-format conversion with integrated buffering

Changes

  • Migrate openai_compat_executor to use RunSSEStream pattern
  • Refactor claudePassthroughProcessor to use IR parser directly
  • Remove 10 unused legacy TranslateXxxResponseStream* functions
  • Remove deprecated types: GeminiCLIStreamState, CodexStreamState

Stats

  • 13 files changed
  • 677 insertions, 1097 deletions
  • Net reduction: -420 lines

v2.0.12

29 Dec 04:48

Choose a tag to compare

Changelog

  • a0197ba fix(streaming): prevent duplicate finish chunks in Claude via Gemini format
  • b5b3825 fix(test): add bun-types reference for process global
  • 9ce4a1a fix(test): declare process type inline
  • 763229b fix(test): update SDK import for OpencodeClient type

v2.0.11

29 Dec 02:52

Choose a tag to compare

Changelog

  • d154020 refactor(translator): simplify IR converters and fix Gemini SDK streaming

v2.0.10

28 Dec 08:31

Choose a tag to compare

Changelog

  • 09af693 fix(translator): preserve thinking signature in Claude streaming responses

v2.0.9

28 Dec 05:36

Choose a tag to compare

Changelog

  • 2eff4ee perf: extend URL pooling and refactor transport for DRY compliance