LiveKit

What is LiveKit?

LiveKit is an open source WebRTC server for real-time voice, video, and data. At its core is a distributed Selective Forwarding Unit (SFU) written in Go that routes live audio and video between participants, paired with client SDKs for the web, iOS, Android, Flutter, React Native, and Rust. On top of that sits an Agents framework for building voice-AI applications.

What is LiveKit best for?

Developers who need to add live audio/video — or a voice-AI agent — to an app and want to own the infrastructure instead of paying per-minute API fees. It fits video conferencing, livestreaming, and telephony, but its standout use case in 2026 is conversational voice AI: OpenAI built ChatGPT’s Advanced Voice Mode on LiveKit.

What can LiveKit do?

• Route live audio, video, and data through a scalable SFU with simulcast, selective subscription, and SVC codecs (VP9, AV1)
• Build voice-AI agents with the Agents framework: automatic turn detection, interruption handling, and a gateway to swap STT, LLM, and TTS providers
• Connect phone calls over SIP telephony, so agents and rooms work across web and the PSTN
• Ship clients fast with SDKs for JavaScript/TypeScript, Swift, Kotlin, Flutter, React Native, and Rust
• Secure sessions with JWT-based access tokens, moderation APIs, and end-to-end encryption
• Deploy as a single Go binary, Docker, or Kubernetes, and scale across multiple regions

Where does LiveKit fall short?

• You bring the AI models. LiveKit orchestrates a voice pipeline but doesn’t ship its own STT, LLM, or TTS — you wire in and pay for providers like Deepgram, OpenAI, or ElevenLabs separately, and their latency and cost become yours.
• Self-hosting the multi-region SFU is non-trivial. A single node is easy, but matching LiveKit Cloud’s global, auto-scaling mesh means running and tuning your own distributed deployment; many teams use the managed cloud precisely to skip this.
• It’s infrastructure, not a finished product. LiveKit gives you a server and SDKs, not a ready-made meeting app — you build the UI, call logic, and recording flows yourself, unlike a drop-in embeddable widget.

Is LiveKit free?

Yes — the LiveKit server, Agents framework, and all client SDKs are free and open source under the Apache-2.0 license, so self-hosting costs only your own servers (roughly $60/month for a couple hundred concurrent users on a modest box). LiveKit Cloud is the paid managed option: a free tier includes 1,000 agent session minutes per month, then usage is billed per track-minute (around $0.004 per audio and $0.006–$0.024 per video track-minute).

What does LiveKit replace?

LiveKit is an open source alternative to proprietary real-time communication APIs — most directly Agora and the video/voice side of Twilio and Vonage. Twilio’s own programmable Video product was discontinued in 2024 (end-of-life December 2026), which has pushed many teams toward self-hostable stacks like LiveKit. Rather than paying metered per-participant-minute rates that scale linearly forever, you run the WebRTC infrastructure yourself and turn it into a largely fixed cost.

FAQ

Is LiveKit open source? Yes. The LiveKit server, Agents framework, and client SDKs are licensed under Apache-2.0, a permissive license with no per-seat or per-minute fees for self-hosting.

Can I self-host LiveKit for free? Yes. Deploy the single Go binary, Docker image, or Kubernetes chart on your own hardware and you pay only for the servers and bandwidth — no LiveKit licensing cost.

Is LiveKit a good Twilio or Agora alternative? For teams that want control over cost and infrastructure, yes — especially for voice AI, where its Agents framework is well ahead of general-purpose video APIs. Managed clouds like Agora still offer more turnkey global infrastructure if you’d rather not operate an SFU.

What do I need to run LiveKit? A server that can run the Go binary, Docker, or Kubernetes, with UDP ports open for WebRTC media and JWT keys configured for authentication. Voice-AI use also needs API access to your chosen STT, LLM, and TTS providers.