Version Packages by github-actions[bot] · Pull Request #1859 · livekit/agents-js

github-actions · 2026-06-23T17:23:07Z

This PR was opened by the Changesets release GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to main, this PR will be updated.

Releases

@livekit/agents@1.4.10

Patch Changes

Add Deepgram keyterm to inference STT model options and support batch recognition in the Deepgram STT plugin. - #1766 (@rosetta-livekit-bot)
chore(dep): update local-inference dep - #1873 (@rosetta-livekit-bot)
Simulation fixes: do not start the worker HTTP server, disable the worker load limit so runs can saturate the agent, and honor LIVEKIT_AGENT_NAME_OVERRIDE so the worker registers under the name lk simulate dispatches to. - #1804 (@u9g)
Emit the lk.chat_ctx span attribute (on LLM-generation and EOU-prediction spans) in snake_case to match the Python framework, which serializes the same attribute via chat_ctx.to_dict(). The JS side was stringifying chatCtx.toJSON() directly, which emits camelCase field names (callId, args, isError, createdAt), so traces from JS agents diverged from Python agents. The chat context is now run through the shared toSnakeCaseDeep conversion (exported from the report layer), and the timestamp exclusion is aligned with Python (toJSON() defaults exclude image/audio/timestamps). Note: the Python EOU path additionally trims to the last N turns and excludes function calls — aligning that behavior is left to a follow-up. - #1804 (@u9g)
Serialize chat-item toJSON() output in snake_case (e.g. new_agent_id, call_id, arguments, is_error, created_at) to match the Python session-report schema. The uploaded chat_history.json previously used camelCase keys, so LiveKit Cloud's parser rejected required fields (e.g. agent_handoff.new_agent_id). Constructor/create() APIs are unchanged. - #1804 (@u9g)
Serialize the session report's events and usage arrays in snake_case to match the Python session-report schema. Event fields (old_state, new_state, is_final, speaker_id, function_calls, created_at, …) and model-usage fields (input_tokens, session_duration, audio_duration, characters_count, total_requests, …) previously leaked camelCase keys, so LiveKit Cloud's Python parser dropped them. Usage durations are now emitted in seconds (matching the proto wire format and Python model). Also adds the audio_recording_path, audio_recording_started_at, sdk_version, and options.user_away_timeout fields to match Python's SessionReport.to_dict(). - #1804 (@u9g)
Serialize the uploaded session-recording chat_history in snake_case to match the Python wire schema. uploadSessionReport was stringifying chatHistory.toJSON() directly, which emits camelCase field names; the snake_case conversion lives only in sessionReportToJSON (toSnakeCaseDeep). As a result the uploaded chat history carried camelCase keys (callId, args, isError, newAgentId, createdAt), and the Python consumer's pydantic validation rejected the chat items with "field required" errors for call_id, arguments, is_error, and new_agent_id. The upload now reuses sessionReportToJSON(report).chat_history so the two serializations can't drift. - #1804 (@u9g)
Keep the STT input timestamp anchor attached to reused STT pipelines across agent handoff. - #1870 (@rosetta-livekit-bot)