# LiveKit Agents — The open-source, self-hostable voice agent framework, Apache 2.0, provider-agnostic, and the infrastructure behind ChatGPT’s Advanced Voice Mode.

> Source: The Agents Index — https://theagentsindex.com/livekit-agents (structured, researched, re-verified)
> Facts last verified: 2026-07-24

LiveKit Agents is an open-source (Apache 2.0) framework, from real-time infrastructure company LiveKit, for building voice AI agents that listen, think and speak in real time. Founded in 2021 by Russ d’Sa and David Zhao as a free, open-source WebRTC project for streaming real-time audio and video, LiveKit evolved into the infrastructure layer behind some of the industry’s highest-profile voice AI products, it powers OpenAI’s ChatGPT Advanced Voice Mode, and its named customers include xAI, Salesforce and Tesla. The company has raised $181.3M total, including a $100M Series C led by Index Ventures (with Salesforce Ventures) in January 2026 that valued it at $1B. Unlike this category’s other three listed platforms, all closed-source, API-metered SaaS you can only consume, LiveKit Agents is a framework you can fully self-host and inspect, with a managed LiveKit Cloud layer available if you’d rather not run the infrastructure yourself.

| Fact | Value |
| --- | --- |
| Website | https://livekit.com/voice-agents |
| Pricing | $0/month |
| API | Yes |
| Best for | Engineering teams that want to own their voice-agent stack end to end — self-hosted or on LiveKit Cloud — without being locked into one vendor’s API. |
| Not for | Non-technical teams who want a hosted dashboard and a working voice agent without writing or deploying code. |

## Verdict

LiveKit Agents is the one genuinely open-source, self-hostable option among this category’s voice agent platforms, an Apache 2.0 framework, not a closed API you can only rent. That openness comes from a company with real weight behind it: a $1B valuation on $181.3M raised, and infrastructure serious enough that OpenAI runs ChatGPT’s Advanced Voice Mode on it alongside customers like xAI, Salesforce and Tesla. The tradeoff is real: unlike Vapi, Retell AI, Bland AI or ElevenLabs Conversational AI, there is no hosted dashboard that gets you a working agent in an afternoon, you write and deploy code, choosing and wiring your own STT/LLM/TTS providers (or LiveKit’s managed Inference gateway) yourself. Pick LiveKit Agents if engineering ownership, self-hosting or avoiding vendor lock-in matter more than a fast, no-code start; skip it if you want a turnkey agent live without writing code.

## Strengths and weaknesses

- ✓ The only open-source, self-hostable framework in this category (Apache 2.0, 11.5k+ GitHub stars), full code visibility and no vendor lock-in, versus three closed-source, API-only rivals.
- ✓ Proven at genuine production scale, not just funding hype: powers OpenAI’s ChatGPT Advanced Voice Mode, and named customers include xAI, Salesforce and Tesla.
- ✓ A genuine free self-serve tier (1,000 agent-session minutes/month, no credit card) despite the company’s $1B valuation, real usage-based pricing published up through a $500/mo Scale tier, not sales-gated from the start.
- ✗ It is a framework and SDK, not a turnkey dashboard builder, you write and deploy agent code yourself, unlike this category’s other three platforms, which configure a working agent through a hosted API/dashboard with far less engineering lift.
- ✗ The free tier’s $2.50 monthly inference credit covers only about 50 minutes of LiveKit-hosted model usage, a real deployment needs either metered LiveKit Inference credits or your own third-party provider API keys billed separately, on top of the agent-session-minute meter, the same stacked-cost pattern this category’s other metered vendors show.
- ⚠ A framework, not a no-code or dashboard-first builder, expect to write and deploy real code, unlike this category’s more configured rivals.
- ⚠ The free tier’s inference credit is small (~50 LiveKit-hosted minutes/month); real deployment cost depends on the LLM/STT/TTS providers and volume you choose, on top of the base agent-session-minute meter.

## Key features

- **Open-source Agents SDK** — Apache 2.0-licensed Python/TypeScript framework (11.5k+ GitHub stars); fully self-hostable, no vendor lock-in.
- **Provider-agnostic pipeline** — Plug in Deepgram, AssemblyAI, OpenAI, Anthropic, ElevenLabs, Cartesia and more, or use LiveKit’s own managed Inference gateway.
- **Built-in real-time plumbing** — Turn detection, interruption handling, background-voice/noise filtering and native SIP telephony out of the box.
- **LiveKit Cloud (optional)** — Managed deployment, autoscaling and observability if you don’t want to run your own infrastructure.

## Use cases

- **Custom enterprise voice agent** — A platform team builds a bespoke phone agent on its own chosen models, self-hosted for full data control.
- **Voice agent inside an existing product** — A team already using LiveKit’s real-time media infrastructure adds a voice agent without adopting a second vendor’s stack.

## Integrations

Speech & LLM providers — Deepgram, AssemblyAI, OpenAI, Anthropic, ElevenLabs, Cartesia and more via plugins, or LiveKit’s own managed Inference gateway · Telephony — native SIP trunking for inbound/outbound PSTN calls · Deployment — self-hosted on your own infrastructure, or LiveKit Cloud (managed) · Model Context Protocol (MCP) — first-class support for connecting agents to external MCP servers over HTTP or stdio

## Sources

- https://livekit.com/voice-agents
- https://livekit.com/pricing
- https://github.com/livekit/agents
- https://livekit.com/blog/introducing-livekit-inference
- https://techcrunch.com/2026/01/22/voice-ai-engine-and-openai-partner-livekit-hits-1b-valuation/
- https://docs.livekit.io/agents/logic/tools/mcp/