agent-voice-bridge

agent-voice-bridge

agent-voice-bridge exposes a phone-friendly web UI and WebSocket bridge for voice conversations with Claude Code.

It runs local speech models on the host machine:

• Silero VAD for voice activity detection
• whisper.cpp via pywhispercpp for speech-to-text
• Kokoro for text-to-speech
• FastAPI + Uvicorn for the web bridge server

Install

pip install agent-voice-bridge

Requirements

• Python 3.10+
• Claude authentication via one of:
  • CLAUDE_CODE_OAUTH_TOKEN
  • ANTHROPIC_API_KEY

Run

agent-voice-bridge

Optional flags:

agent-voice-bridge --host 127.0.0.1 --port 8787 --model sonnet

When the server starts it prints local and network URLs that can be opened on a phone browser.

Release

Build the PyPI artifacts:

npm run voice-bridge:build

Publish to PyPI:

export PYPI_TOKEN=your-token
npm run voice-bridge:publish

Or do both in one step:

export PYPI_TOKEN=your-token
npm run voice-bridge:release

Notes

• The package persists its auth token under ~/.agent-voice-bridge/ by default.
• Set VOICE_BRIDGE_STATE_DIR to override that location.
• Model downloads may happen on first use depending on the underlying speech libraries.