percept-listen
Ambient audio capture and transcription for OpenClaw agents via wearable devices.
What it does
Connects a wearable microphone (Omi pendant or Apple Watch) to your OpenClaw agent. Audio is transcribed locally and streamed as structured conversation data — speaker-tagged, timestamped, and searchable.
When to use
- User wants their agent to hear ambient conversations
- User asks to "start listening" or "turn on the mic"
- User mentions Omi pendant, wearable, or ambient audio
Requirements
- Percept server running locally (
pip install getpercept && percept start) - Omi pendant paired via phone, OR Apple Watch with Percept app
- Webhook configured: Omi app → Settings → Webhooks →
https://<your-tunnel>/webhook/transcript
Setup
# Install Percept
pip install getpercept
# Start the receiver (default port 8900)
percept start
# Or run directly
PYTHONPATH=. python -m uvicorn src.receiver:app --host 0.0.0.0 --port 8900
Configure a tunnel (Cloudflare, ngrok, Tailscale) so Omi can reach your local server.
How it works
- Omi pendant captures audio → phone does STT → sends transcript segments via webhook
- Percept receiver processes segments into conversations
- Conversations are stored in local SQLite with FTS5 full-text search
- All processing stays local — no audio leaves your machine
Data locations
- SQLite DB:
percept/data/percept.db - Live transcript:
/tmp/percept-live.txt - Conversations:
percept/data/conversations/
Configuration
Wake words, speaker names, and all settings are managed via the Percept dashboard (port 8960) or directly in the SQLite database.