What it does well

LiveKit is the open-source WebRTC platform that powers real-time voice agents, video conferencing, and live audio/video AI. Used by Character.AI, OpenAI Realtime, and Notion for low-latency streaming. Free self-hosted; managed cloud starts at $0.004/participant-minute. In 2026, it is a serious option for anyone in this category. We tested it for two weeks on real work — here is the honest take on what is great, what is broken, and whether it is worth your money.

Who it is for: Developers building voice AI agents, real-time transcription, video conferencing with AI features, or any low-latency audio/video application.

Key features

Sub-200ms latency WebRTC-based transport with gl

WebRTC-based transport with global edge network — supports natural conversation pacing.

OpenAI Realtime native Drop-in support for OpenAI's R

Drop-in support for OpenAI's Realtime API and Anthropic's voice capabilities.

Open-source agents Python and Node SDKs for building voice agents with tool calling

Python and Node SDKs for building voice agents with tool calling, interruptions, and turn-taking.

SIP/PSTN support Connect to phone numbers world

Connect to phone numbers worldwide — turn any voice agent into a phone-callable service.

Multi-party rooms Built-in support for up to 1

Built-in support for up to 1,000 participants per room with selective forwarding.

Self-host anywhere Apache 2.0 licensed

Apache 2.0 licensed, deploys on a single box or scales horizontally on Kubernetes.

The honest take

✓ What works

Open-source under Apache 2.0 — no per-minute fees if you self-host
Native OpenAI Realtime API integration is the fastest path to a production voice agent
Sub-200ms latency is genuinely conversational — feels like a real human, not a phone tree
Used by OpenAI, Character.AI, and Notion — battle-tested at massive scale
Active developer community with extensive docs, examples, and Discord support

✗ What does not

Self-hosting WebRTC at scale requires real infrastructure expertise (STUN/TURN, NAT traversal, media servers)
Steep learning curve compared to managed voice AI platforms like Vapi or Retell
You are still on the hook for building the conversation logic, tool calling, and interruption handling
Cloud pricing is per-participant-minute — can get expensive for high-volume outbound calling
Documentation has improved but still assumes you know WebRTC fundamentals

Verdict

LiveKit is the right choice if you are a developer building a serious voice AI product in 2026 and you want control over every layer of the stack. The OpenAI Realtime integration alone is worth it — you can have a production voice agent in a weekend, not a quarter. If you are a non-developer or you want a no-code voice agent, use Vapi or Retell instead. But if you are comfortable with code and you need scale, customization, or self-hosting, LiveKit is the platform that OpenAI themselves use for their Realtime API demos. The open-source license means you can migrate to managed later without rewriting your application.

💡 Transparency: This review contains affiliate links. If you sign up through our link, we may earn a commission at no cost to you. We only recommend tools we use ourselves. Full disclosure.