GitHub – sseanliu/VisionClaw: Real-time AI assistant for Meta Ray-Ban smart glasses — voice + vision + agentic actions via Gemini Live and OpenClaw

If you’ve ever put on smart glasses and thought, “Okay… now what?” you’re not alone. The promise is huge, but the day to day experience can still feel a bit thin. That’s why this project caught my attention.

VisionClaw is an open source experiment that turns Meta Ray-Ban smart glasses into something much closer to a real assistant. Not a demo. Not a gimmick. An assistant that can see what you see, hear what you say, and actually do things for you.

Built by sseanliu and shared on GitHub, VisionClaw combines the Meta Wearables DAT SDK, Google’s Gemini Live API, and an optional layer called OpenClaw. Put the glasses on, tap the AI button, and just talk. The camera streams visual context at about one frame per second, audio flows both ways in real time, and Gemini uses all of that context to respond naturally.

What makes this feel different is the action layer. With OpenClaw connected, the assistant doesn’t stop at answering questions. It can send messages, search the web, manage lists, or even control smart home devices. That moment when an assistant moves from “helpful voice” to “hands actually free” is subtle… and kind of exciting.

There’s also a thoughtful fallback. No glasses? You can run it in iPhone mode and still explore voice and vision features. That’s practical, especially when you’re testing or just getting started.

Setup is very much a builder’s experience. You’ll need a Gemini API key, a bit of Swift file editing, and your devices on the same Wi‑Fi network. Nothing outrageous, but it helps if you’re comfortable poking around configs and logs. I’ve been there, staring at a timeout error, wondering if it’s Wi‑Fi or Bonjour again.

All the source code lives here, and it’s worth browsing even if you’re just curious about where wearable AI is heading:
https://github.com/sseanliu/VisionClaw

VisionClaw feels like a glimpse of what smart glasses are supposed to become. Quietly useful. Context aware. And slowly, step by step, more human to interact with.