BipBip

Vision System
Key workflow
Header

Inspiration

BipBip is inspired by recent notification updates to Claude Code and codex, which demonstrated how helpful basic beeps could be and pushed me to explore further notification options.

What it does

BipBip combines signals from your devices to work out what you're paying attention to right now: whether you're looking at the screen, how far away your phone is, where you are and whether you're on the move. It learns what matters to you through a game-like setup, a conversational voice agent and what it sees you respond to. Then it decides how, and whether, to reach you.

Using the Perceptron API for computer vision, Photon API for rich text notifications and responsive agentic phone calls--to you and an accountability buddy--and careful security measures to work around sensitive agent outputs verified with VEYZL, it works with 15+ of the most common AI agent harnesses precisely to be able to create notifications in the middle of agent execution alongside finished sessions.

Within the notifications themselves, BipBip offers near infinite customization. ElevenLabs can dynamic hear your complaints about notifications, read notifications out loud, and interact with the rich UI entirely hands free. Notifications can be haptic, ultrasonic, and cross-platform, dynamically adapting to surroundings.

How we built it

Agent integrations. One dependency-free Node relay runs as a hook in each agent harness. It maps 16 different hook vocabularies into one event model. An installer edits each agent's config without breaking what's already there: it backs up every file, and disconnecting restores it cleanly. Agent skills (for Claude Code, Codex and the open Agent Skills standard) let an agent connect itself and send mid-task notifications.
Privacy by design. Agents see your code, so we treated their output as sensitive. The relay keeps only an allowlist of fields (event name, session id, project folder, short status codes) before anything leaves the agent's process.
Attention and context. Perceptron for computer vision, Bluetooth signal strength for phone-to-laptop distance, and location for where you are and whether you're driving. Proton for voice agents, message alerts, and hands-free interaction.

Biggest Challenges

The biggest challenge I faced during development was aligning asynchronous technologies towards a central goal. It became unclear at times which technology would best contribute to signal for notifications.

However, combining some outside-the-box thinking from my time researching pose segmentation for rodent behavioral analysis and some extraneous solutions, like streaming meta glass data as a video call where I couldn't get IOS certification, allowed me to define a schema that was fairly satisfactory.

If I had more time, I would love to create a larger suite of features for Meta Glasses. BipBip serves to try to give you more agent information while taking less of your attention.