Skip to content

feat: swap inference abstraction for direct provider plugins, drop ai-coustics#75

Open
NewGenMan wants to merge 1 commit intolivekit-examples:mainfrom
NewGenMan:swap-direct-plugins
Open

feat: swap inference abstraction for direct provider plugins, drop ai-coustics#75
NewGenMan wants to merge 1 commit intolivekit-examples:mainfrom
NewGenMan:swap-direct-plugins

Conversation

@NewGenMan
Copy link
Copy Markdown

Summary

  • Replace inference.STT/LLM/TTS with direct provider plugin calls: deepgram.STT(model=nova-3, language=multi), openai.LLM(model=gpt-5.4), and elevenlabs.TTS()
  • Remove the paid livekit-plugins-ai-coustics dependency and its noise-cancellation room_options block
  • Add turn_handling={interruption: {mode: vad}} to AgentSession to use VAD-based interruption instead of the paid adaptive endpoint
  • Update pyproject.toml extras: livekit-agents[openai,deepgram,elevenlabs,silero,turn-detector]

Why

The inference.* abstraction routes through LiveKit Cloud inference endpoints. Using direct plugins gives full control over model selection and avoids the paid ai-coustics and adaptive-interruption endpoints.

Reviewer notes

  • ELEVENLABS_API_KEY, OPENAI_API_KEY, and DEEPGRAM_API_KEY must be set in .env.local
  • Run uv sync after pulling to install the new plugin extras

Generated with Claude Code

…-coustics

Replace inference.STT/LLM/TTS with direct plugin calls:
- STT: deepgram.STT(model="nova-3", language="multi")
- LLM: openai.LLM(model="gpt-5.4")
- TTS: elevenlabs.TTS()

Update pyproject.toml to add openai/deepgram/elevenlabs extras and
remove livekit-plugins-ai-coustics (paid plugin).

Add turn_handling={"interruption": {"mode": "vad"}} to AgentSession to
use VAD-based interruption and avoid the paid adaptive endpoint.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant