Skip to content

LiveKit Agents 1.8.9 adds decisions through Inference

Updated LiveKit Agents

Summary

A feature-flagged DecisionModel batches named decisions through LiveKit Inference; silent tool progress and new speech providers also arrive.

Category
Voice agents
Date
Pricing
Not stated in the sources we cited.

LiveKit published Agents 1.8.9 on October 24, 2026.

Decisions and tool progress

inference.DecisionModel evaluates named decisions using LiveKit credentials instead of a separate provider key. It batches evaluations and reports a refusal for one decision without discarding the other answers. Access requires a project feature flag; the release does not describe it as universally enabled.

Background tools can now record progress without asking the agent to speak. Their final results can trigger follow-up tool calls. Custom LLM nodes can also attach prompt metadata directly to the provider request's tracing span.

Providers and upgrade cautions

The release introduces a ThunderPhone realtime speech-to-speech plugin and lists additional decision, speech-synthesis and avatar models. Availability remains provider-specific.

There are breaking changes: Rime WebSocket v1 requires an explicit model, and Gradium rejects specifying both voice and voice_id. Inworld connection-pool settings are deprecated and ignored. The notes also list dependency minimums and turn-handling changes, so existing voice applications should review their configured integrations before upgrading.

See the LiveKit Agents hub.

More from LiveKit Agents

View all

Questions

What is LiveKit Agents?

The documentation says the framework lets you add Python or Node.js programs to LiveKit rooms as realtime participants. The repository describes it as a framework for building realtime voice AI agents.

Is LiveKit Agents open source?

Yes. The GitHub API reports Apache-2.0 for livekit/agents.

What is the latest release covered here?

livekit-agents@1.8.11, published November 5, 2026. It adds inference.responses.LLM (REST by default, optional WebSocket), Zoom AI STT, Kalpa TTS, Rime STT and Prism, Fish Audio realtime STT, OpenAI and Gemini live translate models, and moves Respeecher WebSocket API keys into handshake headers.

Where can agents be deployed?

The documentation says agents can be deployed to LiveKit Cloud or any custom environment.

Filed on Agent Radar. Claims in this note are limited to the sources above.