How sk9315/vaani works
This is a LiveKit Agents Python starter for a multi-participant voice assistant, with the Vaani browser client as its local development experience. A person joins a room, speaks or sends text, and the agent routes the turn to a bot for a spoken response; the development client also supports voice selection and preview. The graph includes room context, voice configuration, and the documented model, observability, and audio-cancellation capabilities. Some subsystem links rely on README statements or source-index references; unsampled implementation details are not asserted.
Main entry points: agent.py LiveKit worker · app.js Vaani client · voice_pipeline.py Voice turn processing · dev_server.py Dev token server
Components 17
Browser client
- Vaani client
src/dev_client/app.js - Client presentation
src/dev_client/ui_state.js - Voice orb
src/dev_client/orb.js - Dev token server
src/dev_server.py
Agent runtime
- LiveKit worker
src/agent.py - Voice turn processing
src/voice_pipeline.py
Room coordination
- Turn routing
src/room_coordinator.py - Room context
src/room_context.py - Room UI payloads
src/room_ui.py
Voice services
- Voice profiles
src/voice_config.py
Other components
- Participant
- LiveKit Cloud
- Model inference
- Speech recognition
- Speech synthesis
- Background cancellation
- Agent observability
Connections 20
- Participant → Vaani client: uses
README.md - Vaani client → Client presentation: renders via
src/dev_client/app.js - Vaani client → Voice orb: displays
- Vaani client → Dev token server: requests token
- Participant → Dev token server: requests token
README.md - Vaani client → LiveKit Cloud: joins room
README.md - LiveKit Cloud → LiveKit worker: dispatches session
README.md - LiveKit worker → Turn routing: routes turns
src/agent.py - LiveKit worker → Voice turn processing: processes speech
src/agent.py - Voice turn processing → Turn routing: submits transcript
src/voice_pipeline.py - Turn routing → Room context: records turns
src/room_coordinator.py - LiveKit worker → Room UI payloads: publishes room state
src/agent.py - Room UI payloads → Room context: reads snapshot
src/room_ui.py - LiveKit worker → Voice profiles: configures voices
src/agent.py - Dev token server → Voice profiles: validates and previews
src/dev_server.py - LiveKit worker → Model inference: runs model pipeline
README.md - Voice turn processing → Speech recognition: consumes speech events
src/voice_pipeline.py - LiveKit worker → Speech synthesis: synthesizes replies
README.md - LiveKit worker → Background cancellation: cancels background voice
README.md - LiveKit worker → Agent observability: reports session insights
README.md