LiveKit Agents
LiveKit Agents is an Apache-2.0 framework for building real-time voice agents that run on servers and join a LiveKit room as participants. It connects any speech, language and voice provider, adds semantic turn detection with a transformer model, native MCP tool support and a built-in test framework, and pairs with LiveKit's open-source media server and SIP service, so the entire path can run on your own machines.
What does LiveKit Agents do?
LiveKit Agents "is designed for building realtime, programmable participants that run on servers", for "conversational, multi-modal voice agents that can see, hear, and understand". The agent joins a LiveKit room as a participant, and the framework handles the audio path.
The README lists a mix-and-match ecosystem of speech-to-text, language and text-to-speech providers and realtime models, naming OpenAI, Deepgram, Cartesia and Google in its examples. Semantic turn detection uses a transformer model "to detect when a user is done with their turn", which the README says reduces interruptions. MCP servers plug in "with one line of code". Telephony works through LiveKit's own stack, which includes an open-source SIP service. There is a built-in test framework with judges, and multi-agent handoff appears in the samples.
Key facts
- Licence: Apache-2.0 on the framework. The LiveKit turn detection models are under a separate LiveKit Model License, which the README points to.
- The README's description: "realtime, programmable participants that run on servers" for "conversational, multi-modal voice agents that can see, hear, and understand".
- Providers named in its examples: OpenAI, Deepgram, Cartesia and Google, within a mix-and-match ecosystem of speech, language, voice and realtime models.
- Semantic turn detection uses a transformer model to detect when a person has finished speaking, which the README says reduces interruptions.
- Native MCP support: tools from MCP servers integrate "with one line of code".
- Telephony through LiveKit's own stack, which includes an open-source SIP service.
- Three run modes: console (no external servers), dev and start, the last two connecting to LiveKit Cloud or a self-hosted server through a URL, key and secret.
- Install with pip install "livekit-agents[openai,deepgram,cartesia]"; a Node.js version lives in the separate AgentsJS repository. Latest tagged release when read: livekit-agents 1.8.2.
What does it replace, and where does it fall short?
LiveKit Agents replaces Vapi, Retell AI and ElevenLabs Agents for a team that wants the framework, the media server and the phone bridge all open source on its own machines. The README: "Fully open-source, allowing you to run the entire stack on your own servers, including LiveKit server."
Where it falls short: beyond the console test mode, a LiveKit server is required, either LiveKit Cloud or one you host. The turn-detection models are under a separate LiveKit model licence, which the README points to and this page has not read. The README calls the framework "under active development in a rapidly evolving field".
How does LiveKit Agents run?
pip install "livekit-agents[openai,deepgram,cartesia]" with the providers you need as extras. Three modes: console for a local test with no external servers, dev and start, which connect to a LiveKit server through a URL, key and secret. A Node.js version lives in the separate AgentsJS repository. LiveKit sells LiveKit Cloud and a unified model API called LiveKit Inference; no prices are stated.
Who is LiveKit Agents for?
A team that already runs or is willing to run LiveKit, wants turn detection and testing built in, and wants tools through MCP. A team that wants to avoid running a media server at all should compare Pipecat's transport list.
What limits does the README state?
Active development. Plugin integration tests need credentials and run only for maintainers' pull requests.
Questions people ask
Is LiveKit Agents open source?
Yes. The framework is Apache-2.0, and the README says the whole stack, including the LiveKit media server, is open source and can run on your own servers. One separate piece: the turn-detection models are under a LiveKit Model License, which this site has not read.
Do I need LiveKit Cloud?
No, but you need a LiveKit server. Console mode runs a local test with no external servers; dev and start modes connect to either LiveKit Cloud or a server you host yourself, through a URL, an API key and a secret.
Can LiveKit Agents handle phone calls?
Yes. The README says it works with LiveKit's telephony stack so an agent can make calls to or receive calls from phones, and LiveKit's open-source components include a SIP service. The carrier that provides the number is still yours to arrange.
How does LiveKit Agents compare with Vapi?
It is the framework where the most of the path is open source: the agent, the media server and the SIP bridge. Vapi is a hosted platform with a dashboard and a bill. The alternatives to Vapi page places LiveKit Agents as the pick for running the whole stack on your own servers, beside Pipecat.
Does LiveKit Agents support tools through MCP?
Yes. The README lists native MCP support and says tools provided by MCP servers integrate with one line of code. It also has a built-in test framework with judges, and multi-agent handoff appears in its code samples.
Sources
- LiveKit Agents README and Apache-2.0 LICENSE: github.com/livekit/agents, read 2026-09-16.
Compared with the others
On the open-source alternatives to Vapi page, LiveKit Agents is the pick for the whole stack on your own servers. Apache-2.0. Semantic turn detection, native MCP tool support, a built-in test framework, and a media server plus SIP service that are open source too, so the entire path can run on your machines.
Also on that page: Pipecat for the widest provider and carrier list, TEN Framework for a visual designer, if agora fits, Bolna for a phone agent from a json file.
On the open-source alternatives to Retell AI page, LiveKit Agents is the pick for the whole stack on your own servers. Apache-2.0. Semantic turn detection, native MCP tool support, a built-in test framework, and a media server plus SIP service that are open source too, so the entire path can run on your machines.
Also on that page: Pipecat for the widest provider and carrier list, TEN Framework for a visual designer, if agora fits, Bolna for a phone agent from a json file.
On the open-source alternatives to ElevenLabs Agents page, LiveKit Agents is the pick for the whole stack on your own servers. Apache-2.0. Semantic turn detection, native MCP tool support, a built-in test framework, and a media server plus SIP service that are open source too, so the entire path can run on your machines.
Also on that page: Pipecat for the widest provider and carrier list, TEN Framework for a visual designer, if agora fits.
More voice agent frameworks
A Python framework under BSD-2-Clause, maintained by Daily, for real-time voice and multimodal agents, with the widest provider list on this site and telephony serialisers for Twilio, Telnyx, Plivo, Vonage, Exotel and Genesys.
A real-time conversational AI framework from Agora with a visual designer, a SIP extension and turn detection, released under Apache-2.0 plus conditions that forbid competing deployments, which makes it source available rather than open source.
Added September 16, 2026. Every claim above comes from the project's README, LICENSE or model card, read on September 16, 2026, or from the GitHub API on the date shown in the panel. Found an error? Write to reveneau@licheo.com and it is fixed in the next weekly pass. Repository: github.com/livekit/agents.
Want LiveKit Agents running on your own servers?
Reveneau sets it up on your hardware, connects it to your product, and builds the software around it.