Clawd

The nostalgia angle is real - I grew up with Shimeji and BonziBuddy too. But my honest question is whether the "context-aware" part adds enough to make this a daily driver vs. a novelty you install, smile at for a week, and then disable. The local AI for sentiment analysis is cool in theory, but I'd want to know what the actual CPU/memory overhead looks like on an average laptop with 15+ tabs open. That's usually what kills these ambient extensions for me.

the gemini nano + onnx split is the smart part — nano's gated behind chrome flags and a multi-gig pull on most machines, so the 67mb onnx model is what actually carries the 100%-local promise for everyone else.

The 100% local offline processing is a real differentiator - curious what model is actually running under the hood and whether it performs well on machines without a dedicated GPU?

Running inference entirely local inside a browser extension is a real engineering constraint. Gemini Nano with ONNX Runtime is a smart choice. Context injection from the DOM has to be selective or you'll overwhelm the local model's context window. How do you handle it when the page is large? Does Clawd summarize DOM sections or use a fixed token budget?

The local/offline angle is the part that makes this interesting to me. A browser pet can get creepy fast if page context leaves the machine, so using on-device sentiment and Gemini Nano is a good boundary. The hard product question is making Clawd feel alive without stealing focus. Curious what signals trigger a proactive reaction versus keeping him quiet.

Wait this is actually nostalgic. Desktop pets were the best part of the early internet. The fact that it runs fully local and reacts to what you're browsing is wild. Does it slow down Chrome at all though? My tabs are already fighting for memory.

how heavy is the model to run?are you not afraid that the name is too close to Claude and might be taken down by Anthropic?

the 90s desktop pet nostalgia is real. the fact that it runs 100% local with no data leaving your machine makes it way more trustworthy than it needs to be for a browser pet. the sentiment analysis that reacts to what you're browsing is a fun touch too... curious if it judges me for my 3am wikipedia rabbit holes

What's the actual CPU and memory footprint when Clawd is analyzing pages in the background?

Hi Product Hunt! 👋I’ve always missed the simple joy of desktop companions from the 90s (shoutout to Shimejis and BonziBuddy!), but I wanted something that felt truly "alive" in the modern web.Clawd was born from a simple challenge: Can an AI pet be incredibly smart, highly interactive, and 100% private at the same time?By leveraging Chrome's built-in Gemini Nano alongside on-device machine learning (ONNX Runtime), Clawd "vibe-checks" every page you visit entirely locally. He knows when you’re debugging a crash on GitHub, relaxing on YouTube or doomscrolling on social media—and he reacts to all of it.What makes Clawd special:100% Private AI: Sentiment analysis and chat happen entirely on-device. No API keys, zero latency, and no data leaves your machine.Interactive Voice & Chat: Speak to Clawd via your microphone or text. He replies using text-to-speech voice playback, Markdown formatting, and custom personas (Sarcastic, Gen Z, Encouraging, etc.).Distraction-Free & Optimized: We know your browser footprint matters! Use the Focus Mode to completely silences Clawd or Ghost Mode to dynamically fade Clawd's opacity down while you are actively typing or scrolling, and toggle Performance Mode to cap the physics engine and save battery on low-end machines. Custom Physics: He doesn't just sit there; he crawls the ceiling and snaps to edges.Leveling & Progression: Earn XP simply by browsing to unlock 140+ animations, interactive toys, and rare cosmetic auras.I built this for anyone who spends their entire day in a browser and wants a little more fun and personality in their workspace. I’d love to hear your feedback on his new chat features, or any ideas for new toys and animations to add!Happy browsing!