Percept Listen

What This Skill Does

Percept Listen is a powerful bridge between physical wearable hardware and your OpenClaw agent. It captures ambient audio via an Omi pendant or Apple Watch, transcribes that audio locally, and converts it into structured, timestamped conversation logs. Unlike cloud-based recording solutions, Percept ensures that all processing happens on your machine. The skill integrates directly with a local SQLite database that supports FTS5 full-text search, allowing your agent to recall specific details from physical conversations held throughout the day. It acts as an "always-on" memory supplement for your AI, enabling it to participate in, summarize, or reference context from real-world meetings and discussions.

Installation

To get started, first install the skill via the OpenClaw hub: clawhub install openclaw/skills/skills/jarvis563/percept-listen

Next, you must configure the backend infrastructure. Install the Percept server via pip: pip install getpercept

Once installed, start the receiver process using percept start. You must expose this service to the internet using a tunnel (such as ngrok or Cloudflare) so your Omi pendant or Apple Watch can transmit data to your local machine. Finally, copy your unique webhook URL into the Omi app settings under the 'Webhooks' section to complete the pipeline.

Use Cases

• Meeting Summarization: Automatically generate minutes from in-person meetings without manual note-taking.
• Knowledge Retrieval: Ask your agent, "What was the name of the restaurant we discussed yesterday?" during a real-world dinner.
• Contextual Awareness: Allow your agent to understand physical environment triggers or spontaneous brainstorming sessions.
• Journaling: Use the agent to create transcripts of daily thoughts or interactions for later reflection.

Example Prompts

1. "OpenClaw, start listening. I'm heading into a meeting now."
2. "Search my transcripts for when I talked to Dave about the project deadline last week."
3. "Summarize the last thirty minutes of my conversation and highlight the action items."

Tips & Limitations

Privacy is prioritized, but be mindful of local storage constraints. The SQLite database will grow over time, so ensure you have enough disk space. Because transcription happens locally, the accuracy is dependent on your machine's hardware performance. If you notice delays, ensure your tunnel is configured correctly and that the Percept receiver is not being throttled by background processes. Always be aware of local laws regarding recording private conversations in your jurisdiction.