VisionClaw and the Empty Application Layer of Wearable AI

Every layer of wearable AI has big money in it: hardware, models, agent frameworks. Only the top layer, the applications, is still empty. No product has really defined what it should feel like when an AI is always present once the glasses go on. VisionClaw, from Intent Lab, is the closest attempt to an answer that I have seen in this space. The team is two people. The lead, Xiaoan Liu, is still a PhD student at the University of Colorado Boulder. There is no funding and no pricing, and the website is a single landing page. VisionClaw runs an always-on AI on Meta Ray-Ban glasses. What the camera sees and the microphone hears streams in real time to Gemini. Say "send this to so-and-so" and it carries the task out through OpenClaw, whether that is sending email, looking things up, controlling a smart home, or logging a calendar entry, with more than fifty skills available and no need to touch a phone. Users without glasses can run it with a phone camera. A few technical trade-offs show the thinking. The raw camera feed is 24 fps, and they cut it to one frame a second with image quality halved before feeding the model, trading between latency and accuracy. Audio goes through Gemini Live's native processing, which skips the speech-to-text step and responds faster. The execution layer is left entirely to OpenClaw. The hardware is Meta's, the model is Google's, and the framework is someone else's open source. They only assemble the top layer. The project has 2,600 stars on GitHub. That can't compare with OpenClaw's nearly 200,000, but a fair share of the issues come from people who actually put it on glasses and came back with bug reports and feature requests. In a field where no consumer market exists yet, more than two thousand developers spending time to get it working is a more solid signal than download counts. Intent Lab has two other projects. GlassFlow does real-time transcription and speaker identification for glasses and has 89 stars. Matcha aims to be a voice-first agent operating system for wearables, with 33 stars, and is very early. Together the three make a complete chain from listening to the system to acting on perception.