Skip to content
← Tools / Skills
ToolLiveKit/10 trustVerified
LiveKit logo

LiveKit

Open-source WebRTC stack for realtime voice, video, and AI agents

Visit official site →

About

LiveKit is an open-source platform for realtime communication: a scalable, distributed WebRTC SFU (Selective Forwarding Unit) server written in Go that moves audio, video, and data between people, devices, and AI agents. Around the server sits the LiveKit Agents framework (Python and Node.js) for building code-first voice agents — STT, LLM, TTS, turn detection, tool use, and SIP telephony bundled — plus full client SDKs for web, mobile, desktop, and embedded. People, devices, and agents all join the same room as participants, and agent dispatch routes agents in automatically or on demand.

What it can do

  • Distributed WebRTC SFU: rooms with audio, video, and data tracks; simulcast, selective subscription, speaker detection
  • LiveKit Agents framework: `AgentServer` + `AgentSession` orchestration, intelligent turn detection, interruption handling, tool calling, expressive TTS, keyterm boosting for STT
  • SIP telephony: bridge phone calls in as room participants; dispatch agents into calls
  • End-to-end encryption option, JWT auth, webhooks, data tracks for telemetry, multi-region/distributed deployments
  • Client SDKs for web (JS), iOS, Android, Flutter, React Native, Unity, Rust, embedded C++

Setup

  1. Start a server: self-host with the single binary or Docker (docker run of the livekit-server image), or create a free project on LiveKit Cloud.
  2. Generate API credentials (API key + secret) and create access tokens for clients with the lk CLI or a server SDK.
  3. For voice agents: pip install livekit-agents livekit-plugins-deepgram livekit-plugins-openai livekit-plugins-cartesia (Python 3.10+), export LIVEKIT_URL, LIVEKIT_API_KEY, LIVEKIT_API_SECRET plus model provider keys.
  4. Write agent.py using AgentServer + AgentSession (STT/LLM/TTS wiring), run python agent.py dev, and talk to it in the LiveKit Agents playground.

Prerequisites

Docker or a LiveKit Cloud account for the server; Python 3.10+ (agents) or Node 20+; API keys for your chosen STT/LLM/TTS providers (LiveKit Inference offers models with no per-provider keys); TURN/UDP ports open for self-hosted media.

Use cases

voice AI agentsvideo conferencingtelephony IVRlive streamingrobotics teleoperation

Works with

Claude CodeHermesAntigravityOpenAI Codex

Trust score — 9/10

/10 trustVerified

Scored 9/10 (verified tier, metrics-v1). Evidence: 21,390 GitHub stars (log-scaled vs catalog peers); blended with ~3,609,910 monthly package downloads; last push 0d ago; 23 releases/12mo; 203 open issues. Manual review: 'billions of calls/year' + customer list are README claims; quickstart verified live. License: Apache-2.0.

★ 21,390 GitHub stars

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
8.5/10
Maintenance · recency, releases, issue load
9.6/10
Openness · license + docs
10.0/10
Manual review · sanity read + red-flag check
10.0/10
  • · 21,390 GitHub stars (log-scaled vs catalog peers)
  • · blended with ~3,609,910 monthly package downloads
  • · last push 0d ago; 23 releases/12mo; 203 open issues
  • · manual review: 'billions of calls/year' + customer list are README claims; quickstart verified live

Sources

Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.