This is an early release preview. You may encounter bugs.
LiveKit Agents logo
Unclaimed

Framework voice

LiveKit Agents

Framework for building realtime, server-side voice agents with pluggable STT, LLM, and TTS and WebRTC/telephony support

A+ 96/100 GitHub score ? This grade is derived from GitHub signals, not user votes. Open for the full breakdown.
No votes yet

01 / About

What LiveKit Agents is.

LiveKit Agents is a Python framework for building programmable participants that run on a server and join realtime sessions: voice agents that hear, see, and reply, and text-only agents built with the same code. A Node.js equivalent, AgentsJS, lives in a separate repository.

An Agent holds instructions and function tools; an AgentSession manages the interaction with end users and combines any mix of speech-to-text, LLM, text-to-speech, and realtime API components; an AgentServer schedules jobs and launches agents for user sessions. Model components come either from provider plugins (for example Deepgram, OpenAI, and Cartesia) or from LiveKit Inference, a unified API for models hosted through LiveKit Cloud. Agents connect to LiveKit Cloud or a self-hosted LiveKit server, and a console mode runs an agent locally with microphone and speaker for testing.

Multi-agent workflows hand off between Agent subclasses, with per-agent model overrides and shared typed user data. A test framework runs a session against scripted user input and asserts on the resulting events, including LLM-judged checks on assistant replies.

Features

  • Pluggable models: mix STT, LLM, TTS, and realtime API providers per session, or use LiveKit Inference for hosted models
  • Job dispatch: built-in scheduling and dispatch APIs that connect end users to agents
  • WebRTC clients: LiveKit client SDKs for browsers, mobile, and desktop platforms
  • Telephony: make and receive phone calls through LiveKit's SIP stack
  • Client data exchange: RPCs and data APIs between agents and client apps
  • Semantic turn detection: a transformer model decides when the user has finished speaking
  • MCP support: attach tools from MCP servers to an agent
  • Testing: native pytest integration with event assertions and LLM judges
  • Multi-agent handoff: transfer a session between agents with different instructions and models
  • Video avatars: integrations with Tavus, Bithuman, LemonSlice, and others
  • Coding-agent aids: a LiveKit Docs MCP server and an agent skill for AI coding assistants

02 / Discussion CREDIBILITY-GATED

Discussion

Reading is open to everyone. Posting and voting need a verified identity or a GitHub grade of B or higher.

  • No discussions yet.

04 / Build

Build with LiveKit Agents.

Browse the catalogue for frameworks, tools, and harnesses, each scored on real GitHub credibility.

Get LiveKit Agents →

Browse the catalogue