How kishoreaarya/local_voice-_agent works
Local_Voice-_Agent is a prototype for a local voice assistant: audio is transcribed, a local LLM maps the transcript to a tool payload, and the pipeline looks up that tool for execution. The excerpts also show a separate STT bridge and weather/action helpers, but do not show how they connect to the pipeline. Although the intent prompt lists messaging, calendar, browser, and file tools, their implementations are not present in the sampled runtime. These gaps limit confidence about tool dispatch and several advertised actions.
Main entry points: run_pipeline.py Voice pipeline · agent_brain.py Intent parser · live_agent.py Action executor · stt_bridge.py STT bridge
Components 13
Voice workflow
- Voice pipeline
run_pipeline.py - Whisper CLI
- STT bridge
stt_bridge.py
Intent parsing
- Intent parser
agent_brain.py - Parameter normalizer
agent_brain.py - Local Qwen model
Actions and integrations
- Tool registry
live_agent.py - Action executor
live_agent.py - Weather lookup
live_agent.py - wttr.in
- Speech output
live_agent.py - Platform TTS
Other components
- Voice user
Connections 11
- Voice user → Voice pipeline: provides audio
run_pipeline.py - Voice pipeline → Whisper CLI: transcribes with
run_pipeline.py - Voice pipeline → Intent parser: parses transcript
run_pipeline.py - Intent parser → Local Qwen model: requests completion
agent_brain.py - Intent parser → Parameter normalizer: normalizes parameters
agent_brain.py - Voice pipeline → Tool registry: looks up tool
run_pipeline.py - STT bridge → Whisper CLI: runs transcription
stt_bridge.py - Action executor → Weather lookup: fetches weather
live_agent.py - Weather lookup → wttr.in: requests forecast
live_agent.py - Action executor → Speech output: speaks result
live_agent.py - Speech output → Platform TTS: invokes platform TTS
live_agent.py