Anthropic Claude models in the agent builder

Currently there are models from Google, OpenAI, xAI and others. Are there any plans on integrating the Claude models from Anthropic in the same way soon? They are the most intelligent non-reasoning models and available in Vapi, Elevenlabs Agents etc. so it would be great if they were available in LiveKit too!

@williamBP The livekit-plugins-anthropic plugin already exists, so Claude is available in code today via the Anthropic LLM plugin in your AgentSession. The gap is only in Agent Builder’s UI, not in the runtime.

Whether/when it lands in Agent Builder is a Cloud-team roadmap call only LiveKit can answer. If you can ship code-defined agents, Claude works now; if you specifically need the no-code dashboard surface, this is a ping for @darryncampbell or @CWilson.

Thank you for the input. Yeah I rather meant for the Agent builder UI so I would love to hear if it is planned.

However, with this plugin, do I need my own Anthropic API key or does it go through the universal livekit api like all other llms?

@williamBP, For your second question, the anthropic.LLM(...) plugin uses your own ANTHROPIC_API_KEY (env var or constructor arg) and calls Anthropic directly. You pay Anthropic billing, not LiveKit billing on top.

The dedicated LiveKit Inference LLM list today is DeepSeek, Gemini, Kimi, OpenAI, and xAI, plus an ā€œAdditional modelsā€ bucket. Anthropic isn’t a primary Inference-routed provider yet, so the safe assumption is direct-key only until the Agent Builder UI integration lands, which would presumably ship Inference routing alongside it.

For the roadmap question, only @darryncampbell or @CWilson can confirm.

That is what I suspected. And for an individual its a hassle to deal with the rate limits and all off Anthropic. So using it via a provider like Vapi, Elevenlabs and hopefully Livekit is the only viable option because they can have beneficial enterprise plans that the users can benefit from as well

and also it is great for easy switching of llms in the agent builder

@williamBP the engineering team are aware of the ask to expose Claude through LiveKit Inference, and it is on the roadmap, but it has low priority so I would not expect to see it released as a feature in the near term I’m afraid.

A big shame since they have by far the most intelligent non-reasoning models. Hopefully it changes soon.

But thanks for letting me know!