Prewarm LLM provider connections

[rosetta-livekit-bot]

Summary

• add idempotent, fire-and-forget LLM prewarming with cancellation and close cleanup
• prewarm both the session-configured LLM and the resolved agent-level non-realtime LLM at activity startup
• add provider prewarm implementations for LiveKit Inference, OpenAI, Anthropic, Google Gemini/Vertex, and Mistral
• add deterministic lifecycle, activity-ordering, auth, request-shape, and cancellation coverage
• retain patch changesets for the five directly touched release packages

Python parity

Compared against livekit/agents#6484 and the current Python implementation. JS now matches Python's session-level and activity-level prewarm lifecycle, including Agent.llm overrides, one-shot best-effort behavior, and provider model-list requests. RealtimeModel is excluded from the non-realtime hook, and FallbackAdapter remains intentionally unsupported as in Python.

The JS close lifecycle additionally aborts and awaits in-flight prewarm work and blocks starts after close.

Tests

• strict red-green activity regression proves an agent-level LLM distinct from the session model prewarms exactly once before first inference
• base lifecycle — 9 focused tests for no-op, swallowed rejection, one-shot failure, abort/wait, and close races
• credential-free provider boundaries — Inference token refresh, OpenAI, Anthropic, real-SDK Google Gemini/Vertex auth, and Mistral request/cancellation paths
• full agents suite — 110 files, 1,465 passed and 5 skipped
• Anthropic — 12 passed
• Google — 23 passed after final auth-boundary strengthening
• Mistral — 1 passed, 3 skipped
• OpenAI — 69 passed, 7 skipped
• pnpm typecheck and pnpm build — passed
• changed-file lint, Prettier, IDE diagnostics, and git diff --check — clean
• independent final gates — Spec compliance: APPROVED and Code quality: APPROVED, no findings

API Extractor remains blocked only by documented base-branch namespace-export/missing-baseline issues; declaration builds pass and no new prewarm API warning is introduced.

Cue voice evidence

Fresh exact-head session: sid_915c77d73b3c

The agent-level LLM is distinct from the session LLM and sends a controlled models-list-compatible request to an in-process HTTP boundary. Persisted framework evidence proves:

• exact head 795981ec09298f80618a0b4e035ade1228c9d97d
• exactly one agent-level prewarm request
• event sequence prewarm_request,chat
• resolved predicate agent-prewarm-once-before-chat
• same command reached speaking and persisted the assistant reply
• session and command WAVs are stereo 48 kHz with non-zero samples on both channels

The controlled endpoint proves framework ordering/count and lifecycle dispatch; it does not claim production-provider latency magnitude or TLS reuse.

---

Ported from livekit/agents#6484

[changeset-bot]

🦋 Changeset detected

Latest commit: 795981e

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 39 packages

Name	Type
@livekit/agents	Patch
@livekit/agents-plugin-anthropic	Patch
@livekit/agents-plugin-google	Patch
@livekit/agents-plugin-mistralai	Patch
@livekit/agents-plugin-openai	Patch
@livekit/agents-plugin-anam	Patch
@livekit/agents-plugin-assemblyai	Patch
@livekit/agents-plugin-azure	Patch
@livekit/agents-plugin-baseten	Patch
@livekit/agents-plugin-bey	Patch
@livekit/agents-plugin-cartesia	Patch
@livekit/agents-plugin-cerebras	Patch
@livekit/agents-plugin-deepgram	Patch
@livekit/agents-plugin-did	Patch
@livekit/agents-plugin-elevenlabs	Patch
@livekit/agents-plugin-fishaudio	Patch
@livekit/agents-plugin-hedra	Patch
@livekit/agents-plugin-hume	Patch
@livekit/agents-plugin-inworld	Patch
@livekit/agents-plugin-krisp	Patch
@livekit/agents-plugin-lemonslice	Patch
@livekit/agents-plugin-liveavatar	Patch
@livekit/agents-plugin-livekit	Patch
@livekit/agents-plugin-minimax	Patch
@livekit/agents-plugin-mistral	Patch
@livekit/agents-plugin-neuphonic	Patch
@livekit/agents-plugin-perplexity	Patch
@livekit/agents-plugin-phonic	Patch
@livekit/agents-plugin-protoface	Patch
@livekit/agents-plugin-resemble	Patch
@livekit/agents-plugin-rime	Patch
@livekit/agents-plugin-runway	Patch
@livekit/agents-plugin-sarvam	Patch
@livekit/agents-plugin-silero	Patch
@livekit/agents-plugin-soniox	Patch
@livekit/agents-plugin-tavus	Patch
@livekit/agents-plugins-test	Patch
@livekit/agents-plugin-trugen	Patch
@livekit/agents-plugin-xai	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no potential bugs to report.

View in Devin Review to see 1 additional finding.

[toubatbrian]

Completed Rosetta parity and verification at exact head 795981ec09298f80618a0b4e035ade1228c9d97d.

The resolved branch adds the missing agent-level activity prewarm path, close-race protection, deterministic base/provider coverage including real-SDK Vertex auth, and fresh Cue voice evidence sid_915c77d73b3c proving one controlled prewarm request before first chat with distinct session/agent LLMs. Full focused/provider/agents tests, typecheck, build, formatting, lint, stereo WAV checks, independent spec/quality review, and exact-head CI are green. Zero unresolved threads; mergeable.