Skip to content

Latest commit

 

History

History
52 lines (40 loc) · 3.2 KB

CHANGELOG.md

File metadata and controls

52 lines (40 loc) · 3.2 KB

Changelog

v0.2.1 - 2025-12-06

Changed

  • If LM Studio is not running, only show a warning about that when LLM_LMSTUDIO_DEBUG=1. This allows having the plugin installed even if you do not use LM Studio models all the time.

v0.2.0 - 2025-12-03

Added

  • Full tool-calling support in both sync and async flows, including encoding of prompt.tools, replaying prior tool results, and surfacing streamed tool invocations back through llm.Response.
  • shell.nix based setup of development environment
  • End to end tests to validate plugin.

Changed

  • All discovered LM Studio models are now registered as lmstudio/<model_id> (or lmstudio@host/... when multiple servers are configured) so CLI usage is consistent across environments.
  • Migrated the package to a src/ layout, added an explicit [build-system] table, generated a pinned uv.lock, refreshed README instructions to favor uv run --all-extras pytest.

Fixed

  • Automatic lms load calls now pass --exact, --host, --port from LMSTUDIO_API_BASE, and optional --ttl $LLM_LMSTUDIO_TTL, show clearer progress, and use longer timeouts to make cold starts and long responses more reliable.

Breaking Changes

v0.1.0 - 2025-05-07

Added

  • Initial support for Vision Language Models (VLMs) allowing image attachments with prompts via llm -a path/to/image.png ....
  • Plugin automatically detects and registers models from local LM Studio server(s).
  • Supports LM Studio API /v1 and newer /api/v0 for model discovery and interaction.
  • LMStudioModel and LMStudioAsyncModel now include a display_suffix (e.g., "● 👁️ ⚒️") to indicate loaded status, vision support, and schema/tool support, visible in llm inspect <model_id>.
  • Asynchronous operations and testing using pytest-asyncio and pytest-vcr.
  • VCR cassettes for asynchronous tests to enable CI and offline testing.
  • LMSTUDIO_API_BASE environment variable can now take a comma-separated list of server URLs.
  • Automatic model loading attempt via lms load <model_id> if the target model isn't loaded (requires lms CLI).
  • Debug logging controlled by LLM_LMSTUDIO_DEBUG=1.
  • Support for embedding models.

Fixed

  • Correct construction of OpenAI-compatible payloads for messages, including base64 encoded images.
  • Handling of system prompts for both current turn and conversation history.
  • Resolved various test failures to achieve a stable test suite.
  • Ensured pytest-vcr correctly records and replays interactions for async tests.
  • Model ID handling for reliable lookup (llm -m <model_id>).
  • llm.Response.text() is now correctly awaited in async tests.

Changed

  • Updated README with improved installation instructions, usage examples, vision support details, configuration, and a development section including VCR cassette generation.

Known Issues

  • Icons in display_suffix may not render correctly in all terminals when using llm models list but are usually visible with llm inspect.