LiveKit Agents
Python framework for realtime voice and multimodal AI agents that join LiveKit rooms as participants.
Programmable agent participants with STT, LLM, TTS, telephony, turn detection and MCP support.
LiveKit Agents is a Python framework for building realtime, programmable participants that run on servers and can see, hear and speak. It suits developers who want voice agents on WebRTC with built-in job dispatch and telephony.
What it is
You define an Agent with instructions and tools, wrap it in an AgentSession with STT, LLM, TTS or a realtime API, and run it as a worker that joins a LiveKit room. The framework adds dispatch APIs, semantic turn detection, RPC and data exchange with clients, native MCP support and a built-in test framework with judges.
Who it's for
It suits developers building production voice agents for support lines, assistants or phone calls who are fine using LiveKit's stack. It does not suit non-coders, and the best-integrated path uses LiveKit Cloud services that are billed.
Key features
Agent sessions
Agent, AgentSession and entrypoint objects handle conversations with end users.
Provider mix and match
Combine STT, LLM, TTS and realtime APIs from many providers, or use LiveKit Inference.
Job scheduling
Built-in task distribution with dispatch APIs connects end users to agents.
Semantic turn detection
A transformer model detects when a user has finished speaking to reduce interruptions.
Telephony
Works with LiveKit's SIP stack so agents place and receive phone calls.
MCP and test framework
Native MCP tool support, plus a built-in test framework that uses judges to check agent behaviour.
Up and running in 3 steps
Our honest verdict
What's great
- Realtime WebRTC and telephony come from the same ecosystem, with open-source client SDKs for all major platforms.
- You can run the whole stack on your own servers, including the open-source LiveKit media server.
- Semantic turn detection and job dispatch are built into the framework.
- Release notes show frequent updates; 1.8.11 added Responses API support and new providers.
Where it falls short
- You need a LiveKit server or LiveKit Cloud, plus API keys for every model provider, on top of pip install.
- Examples default to LiveKit Inference through LiveKit Cloud, whose plans start at $50/mo (Ship) beyond the free Build tier.
- The LiveKit turn detection models are under a separate LiveKit Model License, not Apache-2.0.
- The README itself says the framework is under active development in a rapidly evolving field, so APIs change.
Pricing & license
The framework is free under Apache-2.0. livekit.io/pricing shows LiveKit Cloud with a free Build plan at $0/mo, Ship from $50/mo and Scale from $500/mo, with Enterprise on request; self-hosting costs only your own servers and model APIs.
Open remote AI-training roles
767 open remote AI-training roles, updated November 14, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Can I self-host everything?
The README says the entire stack can run on your own servers, including the LiveKit server.
Is there a JavaScript version?
Yes, the README points to AgentsJS for TypeScript.
What licence applies to the turn-detection models?
The LiveKit Model License, separate from the Apache-2.0 framework.
