The open-source network that lets AI agents see, hear, and speak in real time - the plumbing behind ChatGPT's voice.
When 800 million people speak to ChatGPT each week and hear it answer back in a natural, conversational voice, they are - without knowing it - using LiveKit. The San Francisco company builds the real-time layer that carries audio, video, and data between people and machines, fast enough that a spoken exchange feels like a conversation rather than a walkie-talkie call.
LiveKit started in 2021, when the world had moved onto video calls and co-founders Russ d'Sa and David Zhao set out to build an open-source stack for WebRTC, the browser standard for real-time media. The first product was a media server for livestreaming and conferencing. It was useful, but modest. Then large language models learned to talk, and the modest media server turned out to be exactly the piece the AI industry was missing.
The problem LiveKit solves is deceptively narrow: moving media between many participants with very low latency, reliably, at global scale. Foundation models can now generate speech and understand it. Applications can be built on top. But between the model and the user sits a hard networking problem - routing audio across continents, handling packet loss, detecting when a speaker has finished a turn, and letting a user interrupt an agent mid-sentence. That middle layer is what LiveKit owns.
Its answer has two parts. The open-source framework, including LiveKit Agents, is free to download and self-host, which is how the company reached more than 300,000 developers. LiveKit Cloud is the managed version: a global edge network with region pinning that keeps media close to users, billed per concurrent session. The core media server is written in Go; performance-critical paths lean on Rust.
The customers are a roll call of the AI era. OpenAI uses LiveKit for ChatGPT's Advanced Voice Mode. xAI, Salesforce's Agentforce, Tesla, Meta, and Spotify are all named users, and the same technology backs some 911 emergency operator systems and mental-health providers. Collectively, LiveKit says, its network facilitates billions of calls each year.
In January 2026, that traction was priced. LiveKit raised a $100 million Series C led by Index Ventures, with Salesforce Ventures, Altimeter Capital, Redpoint Ventures, and Hanabi Capital participating, at a $1 billion valuation. The company had raised roughly $83 million across its earlier rounds, bringing the total to about $183 million. The pitch to investors was less about today's voice apps than about the shape of the next decade: a voice-driven era of computing, and the runtime it will run on.
Provides open infrastructure - a media server, SDKs, an agents framework, and a managed edge network - for building real-time voice, video, and physical AI applications.
Over 300,000 developers, from solo builders to AI labs and Fortune 500s: OpenAI, xAI, Salesforce, Tesla, Meta, Spotify, Coursera, and emergency services.
Real-time media routing is hard: latency, packet loss, turn detection, interruption handling, global scale. LiveKit makes it a dependency, not a project.
Open source and self-hostable, with a low-latency edge network the company owns - versus closed, managed rivals like Vapi, Retell, Twilio, and Agora.
From an open-source media server to a full platform for voice, video, and physical AI agents.
A scalable WebRTC selective forwarding unit in Go, for real-time audio, video, and data.
Build custom voice and multimodal agents with turn detection, interruption handling, and session management.
Global edge network with region pinning and low-latency routing, billed per concurrent session.
Web, iOS, Android, React Native, and backend languages for real-time media.
Model routing across STT, TTS, and LLM providers for voice pipelines.
Session replays, traces, and transcripts for debugging deployed agents.
Carrier integrations that bring voice agents onto phone networks.
Named users span the biggest AI labs and Fortune 500s - collectively facilitating billions of calls a year.
| Round | Amount | Date | Lead / Investors |
|---|---|---|---|
| Seed | within ~$83M pre-C | 2021 | Redpoint Ventures; angels incl. Jeff Dean, Elad Gil |
| Series A / B | within ~$83M pre-C | 2022–2023 | Altimeter Capital, Redpoint Ventures |
| Series C | $100M | Jan 2026 | Index Ventures (lead), Salesforce Ventures, Altimeter, Redpoint, Hanabi |
Co-founder & CEO. Previously co-founded Evie Labs, acquired by Medium in 2019, and worked at Twitter and 23andMe. Goes by @dsa on Twitter and GitHub. Leads LiveKit's push toward voice-driven computing.
Co-founder & CTO. Co-created LiveKit's open-source WebRTC stack in 2021 and leads the engineering behind its media server and global edge network.
Russ d'Sa and David Zhao launch an open-source WebRTC media server as the world shifts to video calls.
A managed global network makes the open-source stack easy to scale.
A framework for building voice and multimodal AI agents.
LiveKit becomes the real-time layer behind ChatGPT's Advanced Voice Mode.
Index Ventures leads a round to expand compute, storage, and network for voice computing.
Open-core. The media server and Agents framework are free to self-host; LiveKit Cloud is usage-based - around $0.01/min per concurrent AI agent session, plus per-component fees for STT, TTS, telephony, and observability. Enterprise contracts add the rest.
The infrastructure layer between foundation models and end-user apps. It competes with managed voice platforms (Vapi, Retell, Bland), real-time media APIs (Agora, Daily, Twilio, 100ms), and open frameworks (Pipecat).
It provides open-source software and a global edge network for building real-time voice, video, and AI agent applications with very low latency.
Yes. The core media server and Agents framework are open source and free to self-host. LiveKit Cloud is a paid, usage-based managed version.
Over 300,000 developers, plus companies like OpenAI, xAI, Salesforce, Tesla, and Spotify. It powers ChatGPT's Advanced Voice Mode.
About $183M total, including a $100M Series C in January 2026 led by Index Ventures at a $1 billion valuation.
Russ d'Sa (CEO) and David Zhao (CTO) founded the company in 2021 in San Francisco.