LiveKit Agents
Framework for building real-time voice and multimodal AI agents over WebRTC.
LiveKit ships a v1.0 turn detector, its clearest move on voice-agent latency
◆Recent moves
- 1d ago
@livekit/agents@1.6.5
A fix release broadening Gemini 3 Flash model detection, handling AssemblyAI mode presets, and improving worker retry behavior. Routine maintenance.
View source ↗ - 16d ago
livekit-agents@1.6.4
Disables retries on end-of-turn connection errors and updates dependencies and examples. Reliability tuning, no user-facing feature.
View source ↗ - 18d ago
livekit-agents@1.6.3
Fixes across speed options, realtime tool_choice handling, and end-of-turn timeout behavior. Bug-fix release.
View source ↗ - 21d ago
livekit-agents@1.6.2
Adds AssemblyAI universal-3-5-pro as the new default STT with Voice Focus streaming params, plus Gemini 3.1 Flash TTS support. Real broadened model coverage, consistent with the provider-integration arc.
View source ↗ - 21d ago
livekit-agents@1.6.1
⚡ SPARKIntroduces LiveKit Turn Detector v1.0, a model that combines audio and text semantics to decide the optimal moment for the agent to respond. It targets the core turn-taking problem in voice agents and is the most directional release in the recent line.
View source ↗ - 29d ago
livekit-agents@1.6.0
Adds Asynchronous Tools, letting a long-running tool hand control back to the LLM and stream progress into the conversation instead of leaving the user in silence. A meaningful responsiveness primitive that complements the turn-detection work.
View source ↗