LiveKit Agents v1.8.11 runs the Responses API on LiveKit Inference and adds live translation

LiveKit Agents v1.8.11 runs the Responses API on your LiveKit Inference credentials, with structured output and web search. Turn on the WebSocket transport and one connection carries every turn of the call.

Slide text

LiveKit Agents Added

LiveKit Agents v1.8.11 runs the Responses API on LiveKit Inference

llm = inference.responses.LLM("openai/gpt-4.1", use_websocket=True)

Structured output and web search · REST by default, WebSocket opt-in

GitHub release · livekit-agents@1.8.11 · Responses API through LiveKit Inference

Zoom Scribe brings speech-to-text that streams live or works one utterance at a time. Fish Audio adds realtime transcription with word timings, and Kalpa adds streaming speech in English and Hindi.

Slide text

LiveKit Agents Added

Zoom, Fish Audio and Kalpa join as speech providers

Zoom AI Zoom Scribe STT · live streaming and per-utterance

Fish Audio transcribe-stream-1 realtime STT · word timings

Kalpa kalpa-tts-1 streaming TTS · English and Hindi

GitHub release · livekit-agents@1.8.11 · New providers and models

Rime now covers both ends of a call: speech-to-text, and Prism, a speech-to-speech model behind the standard realtime interface. Prism also runs on LiveKit Inference. Both need Python 3.11 or later.

Slide text

LiveKit Agents Added

Rime adds speech-to-text and Prism speech-to-speech

Rime STT Speech-to-text plugin

Prism Speech-to-speech through the standard realtime interface

Inference rime/prism joins the LiveKit Inference realtime models

Both require Python 3.11 or later · STT ships as an extra

GitHub release · livekit-agents@1.8.11 · New providers and models · #8396

Your agent can now translate a call as it happens, in beta. OpenAI's GPT Realtime Translate transcribes with GPT Realtime Whisper, and Google's Gemini 3.5 Live Translate runs on Gemini or Vertex AI.

Slide text

LiveKit Agents Added

OpenAI and Google bring live translation to your agent

OpenAI gpt-realtime-translate · gpt-realtime-whisper transcription

Google gemini-3.5-live-translate-preview · Gemini API and Vertex AI

Both ship through the beta translation interface

GitHub release · livekit-agents@1.8.11 · New providers and models

Inference Responses now starts each known model at its lowest reasoning effort. A late transcript joins the turn the user resumed, and uninterruptible speech keeps what the user said without stopping or replying.

Slide text

LiveKit Agents Changed

Your agent reasons less by default and keeps more of what users say

Responses Each known model starts at its lowest reasoning effort

STT turns A late final joins the turn the user resumed

Uninterruptible User transcripts kept · no interruption, no reply

The provider’s reasoning default is still one setting away

GitHub release · livekit-agents@1.8.11 · Behavior changes

Respeecher now sends your API key in the WebSocket handshake instead of the request URL, and adds Swedish, Brazilian Portuguese, German and Arabic models. If you run Respeecher, I would upgrade today.

Slide text

LiveKit Agents Security

Respeecher keeps your API key out of request URLs

Auth WebSocket API keys move to handshake headers

Models Swedish, Brazilian Portuguese, German and Arabic TTS

LiveKit Agents v1.8.11 release notes on GitHub

GitHub release · livekit-agents@1.8.11 · Security