LiveKit

Open-source real-time infrastructure for production voice and video AI agents

LiveKit is a San Jose-based open-source platform for building, deploying, and scaling real-time voice and video AI agents, built on top of WebRTC with a managed selective forwarding unit (SFU) that handles signaling, media routing, and scale — so developers focus on agent logic rather than infrastructure. It is the de facto infrastructure layer for voice AI in 2025–2026: OpenAI runs ChatGPT’s Advanced Voice feature on LiveKit’s infrastructure, serving millions of voice conversations daily. The LiveKit Agents framework (went 1.0 in April 2025, currently on Python 1.5.x) provides a production-ready orchestration layer for building STT → LLM → TTS pipelines with support for multi-agent architectures, native SIP telephony, MCP tool integration, and semantic turn detection. Unlike no-code managed voice platforms, LiveKit is developer-first and gives engineering teams full control over every component of the voice pipeline while providing LiveKit Cloud as a managed hosting option. At scale, the framework path undercuts managed platforms by 60–80% per call cost above roughly 10,000 minutes per month.

Compliance

GDPR HIPAA SOC 2

Visit LiveKit

Key Features

LiveKit Agents framework (open-source): A Python (and Node.js) orchestration layer for voice and multimodal AI agents that manages the full STT → LLM → TTS pipeline lifecycle — with pluggable components so teams choose their own models (Deepgram, AssemblyAI, OpenAI, Cartesia, ElevenLabs, and many others) without lock-in.
Semantic turn detection: A transformer model that detects when a user is done with their turn, reducing interruptions compared to standard voice activity detection (VAD) approaches — a critical differentiator for natural conversational agents.
Native SIP telephony: LiveKit SIP and Phone Numbers shipped GA in 2025, allowing voice agents to accept a call from any phone on Earth with roughly four lines of configuration — removing the need for a separate Twilio bridge.
LiveKit Inference: A managed inference gateway co-located with LiveKit’s SFU, providing unified access to STT, LLM, and TTS models from OpenAI, Deepgram, Cartesia, ElevenLabs, and more through a single billing relationship — designed to minimize cross-region latency.
MCP tool support: Native support for MCP that lets developers integrate tools provided by MCP servers with one line of code, enabling voice agents to take real-world actions across connected business systems.
Fully self-hostable: The entire stack — LiveKit server, Agents framework, and SFU — is open-source under Apache 2.0 and can be run on the customer’s own infrastructure, critical for HIPAA, GDPR, and data-sovereignty requirements.

Use Cases

For teams building production voice AI agents: An engineering team uses LiveKit Agents as the orchestration layer for a customer support voice agent, assembling their own choice of STT, LLM, and TTS components without paying per-minute managed platform fees.
For inbound and outbound telephony automation: A company uses LiveKit’s native SIP integration to deploy agents that handle inbound support calls and outbound qualification calls on real phone numbers without a separate telephony provider.
For regulated industries requiring self-hosted infrastructure: A healthcare or financial services company self-hosts the full LiveKit stack to keep all voice data within its own infrastructure, meeting HIPAA and data-sovereignty requirements.

Pricing MODELS

Freemium

Pricing Summary

Free to self-host (Apache 2.0). LiveKit Cloud (managed hosting) is usage-based, starting free for development; production pricing scales with minutes of audio and active sessions. Enterprise is custom-quoted with dedicated support and SLA guarantees. LiveKit Inference (model access via LiveKit) is billed separately by usage.