Ambient audio capture and local transcription for AI agents via wearable devices like Omi and Apple Watch.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install percept-listen
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install percept-listen using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Percept Listen?
Percept Listen is a technical bridge that connects wearable audio hardware directly to your local environment, designed specifically to enhance the capabilities of Openclaw Skills. By using an Omi pendant or an Apple Watch, this skill captures ambient conversations and streams them as structured, speaker-tagged data. This allows your agent to maintain a continuous context of the physical world, making it an essential tool for developers building proactive digital assistants.
The core value of this integration for Openclaw Skills lies in its privacy-first approach. While speech-to-text is performed on the mobile device, all transcript processing, storage, and indexing occur on your local machine. This ensures that sensitive conversation data remains under your control while providing a searchable, time-stamped record of every interaction recorded by your wearable devices.
Percept Listen Use Cases
- Enabling an agent to monitor and log live ambient conversations during meetings or brainstorming sessions.
- Activating a passive listening mode for your Openclaw Skills via triggers like "start listening" or "turn on the mic."
- Integrating wearable hardware such as the Omi pendant or Apple Watch into a unified, local AI research workflow.
- Maintaining a searchable digital memory of offline discussions that can be indexed and queried by Openclaw Skills.
How Percept Listen Works
- The wearable device (Omi pendant or Apple Watch) captures audio and uses the mobile app for initial speech-to-text processing.
- The smartphone sends transcript segments via a configured webhook to the local Percept server.
- The Percept receiver collects these segments and organizes them into coherent, speaker-tagged conversations.
- All processed data is stored in a local SQLite database, utilizing FTS5 for high-performance full-text search capabilities.
- Your Openclaw Skills can then access the local database or live text files to retrieve the context of real-world conversations in real-time.
Percept Listen Setup
To integrate this capability with your Openclaw Skills, start by installing the Percept package:
pip install getpercept
Launch the local receiver server to begin listening for incoming webhooks:
percept start
Alternatively, you can run the receiver using uvicorn directly:
PYTHONPATH=. python -m uvicorn src.receiver:app --host 0.0.0.0 --port 8900
You must also configure a tunnel (such as Cloudflare, ngrok, or Tailscale) to allow the Omi app to reach your local server. Point the Omi app webhook settings to https://<your-tunnel>/webhook/transcript to finalize the connection to your Openclaw Skills ecosystem.
Percept Listen Data Schema & Taxonomy
The skill manages its data locally to ensure speed and privacy for all Openclaw Skills interactions. The data structure is organized as follows:
| Data Point | Path | Purpose |
|---|---|---|
| SQLite DB | percept/data/percept.db |
Stores all indexed transcripts and metadata using FTS5. |
| Live Transcript | /tmp/percept-live.txt |
A temporary file for real-time monitoring of current audio. |
| Conversation Logs | percept/data/conversations/ |
Directory containing individual session records for Openclaw Skills retrieval. |
All metadata, including speaker identification and wake words, is accessible via the Percept dashboard on port 8960.
Percept Listen Advanced Features
- Support for multiple wearable hardware inputs, including Omi pendants and Apple Watch apps.
- Local-only data processing to maintain strict privacy for all conversations recorded via Openclaw Skills.
- Full-text search integration using SQLite FTS5 for instant retrieval of historical conversation context.
- Custom dashboard for managing speaker profiles, wake words, and database configurations.
- Live file-streaming capabilities that allow other Openclaw Skills to react to speech segments as they are generated.
SKILL.md
Loading
Related Openclaw Skills
percept-ambient
An always-on ambient intelligence layer that builds a continuous knowledge graph of your conversations and projects without needing explicit commands.
Browser Audio Capture
A bridge that enables AI agents to capture and process real-time audio streams from any browser tab without platform-specific integrations.
watchclaw
An automated health monitoring and recovery watchdog designed to protect OpenClaw gateways from faulty configuration changes.
ShieldCortex
ShieldCortex provides AI agents with a secure, persistent memory system featuring semantic search, knowledge graphs, and a 6-layer defense pipeline.
Percept Speaker ID
A sophisticated speaker identification system that maps anonymous transcript labels to real names and manages voice command authorization levels.
percept-summarize
An automated conversation intelligence tool that generates AI summaries and maps entity relationships from transcribed voice data.
Featured*
