Skip to content

CHANGELOG.md

Latest commit

 

History

History
120 lines (94 loc) · 5.34 KB

CHANGELOG.md

File metadata and controls

120 lines (94 loc) · 5.34 KB
Raw

Changelog

All notable changes to this project are documented here.

The format follows Keep a Changelog, and the project follows Semantic Versioning.

Unreleased

1.0.0-rc.2 - 2026-06-21

Added

  • Runtime validation for LM Studio's native GET /api/v1/models schema.
  • Effective context tests for unloaded, singly loaded, and multiply loaded model instances.
  • Issue-shaped Nemotron, GLM vision, authentication, changing-model-list, embedding, and context regressions.
  • A real OpenCode ACP SDK gate for initialization, session creation, and newline-delimited JSON-RPC stdout purity.
  • A clean HOME/XDG test for OpenCode's native npm-plugin resolver and cache.
  • Fixture provenance and a documented v1 endpoint/mapping contract.

Changed

  • The v1 line now requires LM Studio 0.4.0 or newer and OpenCode 1.17.7 or newer.
  • Discovery uses only native /api/v1/models; OpenAI-compatible /v1/chat/completions remains the inference boundary.
  • Model key and display_name replace v0 IDs and name heuristics.
  • Vision support comes from capabilities.vision.
  • Loaded models use their configured instance context. Multiple active instances use the conservative minimum; unloaded models use the model maximum.
  • Automatic discovery checks only LM Studio's documented default port. Custom ports remain explicit provider configuration.
  • The pinned TUI compatibility gate now runs the supported OpenCode 1.17.7 minimum, while current CLI/ACP checks run 1.17.9.

Fixed

  • Generated model entries and whitelists no longer retain IDs removed by a later LM Studio configuration load.
  • Loaded and unloaded embedding records remain outside the chat provider.
  • HTTP-200 error objects are rejected as unsupported native responses.
  • A false trained_for_tool_use value no longer risks being interpreted as a reason to disable OpenCode tools.

Compatibility notes

  • Stable 0.3.1 remains available for earlier installations.
  • trained_for_tool_use and LM Studio reasoning settings are not translated into non-equivalent OpenCode fields. See docs/v1-contract.md.
  • The prerelease remains on npm next; stable npm latest remains 0.3.1.

1.0.0-rc.1 - 2026-06-21

Added

  • Runtime-validated LM Studio /api/v0/models discovery using an explicit typed boundary.
  • Official model-type-driven chat, vision, embedding filtering, and context metadata.
  • Structured OpenCode logging through client.app.log.
  • Real OpenCode CLI smoke coverage and Microsoft TUI Test end-to-end coverage.
  • Deterministic, generated terminal screenshots for README and release evidence.
  • Release preflight, clean package inspection, and dependency audit gates.

Changed

  • Model discovery uses LM Studio's reported model type and metadata.
  • User model overrides and explicit whitelists take precedence over discovery.
  • Context-aware models use LM Studio's reported context length and a bounded output reserve so OpenCode compaction retains a usable prompt budget.
  • Auto-detection accepts a port only after its LM Studio metadata response validates.
  • Package builds contain a fresh dist/ tree generated from current source.
  • OpenCode configuration validation delegates to OpenCode's own parser.

Fixed

  • Embedding models remain outside OpenCode's chat-model configuration.
  • Explicit and private-network API-key authentication applies during discovery.
  • Vision models receive the OpenCode attachment capability only when LM Studio reports vlm.
  • Downloaded models remain eligible for LM Studio JIT loading.

Release-candidate scope

The proposed 1.0.0 release defines a typed integration contract between LM Studio's REST metadata and OpenCode's config hook. This first candidate is published under npm next; stable npm latest remains on 0.3.1 while community feedback is collected in #34.

Migration notes

  • LM Studio 0.3.6 or newer is required for /api/v0/models discovery.
  • Model discovery uses GET /api/v0/models.
  • Model IDs are preserved exactly.
  • OpenCode's chat provider includes llm and vlm model types.
  • Users needing a custom display name, output limit, or whitelist should configure an explicit model override.
  • Final 1.0.0 will not occur until RC feedback is triaged and CI, current-OpenCode smoke tests, pinned screenshot tests, package inspection, PR disposition, and explicit approval are complete.

Acknowledgements