Highlighting to this channel, if you missed the an...
# agents
l
Highlighting to this channel, if you missed the announcement of Agent Builder last week, you can find the blog post at https://blog.livekit.io/announcing-agent-builder-a-pathway-to-code/. Now you can create, test and deploy production-grade agents all within your browser.
a
@lively-winter-88668: Tried it wanted to know when are you guys planning to add STT models for Hindi language? Also the turn taking is not optimized that's what I felt, is it the traditional VAD that is being used here?
l
The intention is to support the same STT providers in agent builder as we support in LiveKit Inference, i.e. AssemblyAI, Cartesia, and Deepgram. Once you have your agent built and working with Agent Builder, you would: • Download the code • Swap out the stt for a LiveKit plugin that supports Hindi, for example https://docs.livekit.io/agents/models/stt/plugins/sarvam/ • Deploy your updated agent For turn taking, this is using our Multilingual model,
turn_detection=MultilingualModel(),
, so should be performant - that is the model described at https://docs.livekit.io/agents/build/turns/turn-detector/ (it also supports Hindi).
s
@lively-winter-88668 can i use my own tts solution with the agent builder?
l
@salmon-book-23491 specifically in agent builder? No, the idea would be to use agent builder to create your agent as far as possible, even if that's just the boiler plate, then download the code and swap out for your own TTS as described above. I'm not sure specifically what you mean by your "own tts solution", but you could run your own model on Baseten, https://docs.livekit.io/agents/models/stt/plugins/baseten/, or create your own livekit plugin, https://github.com/livekit/agents/tree/main/livekit-plugins... but neither of those can be done within agent builder.
s
@lively-winter-88668 thank you! So basically, the agentbuilder does not allow to use plugins - we have a own tts plugin that serves our own tts models
✅ 1