Browser AI Chat
AI chat application that runs entirely in the browser using local models like Gemini Nano, Transformers.js, and WebLLM for private, serverless AI.
About
An AI chat application that runs entirely in the browser — no server, no API keys, no data leaving your device. It supports multiple local AI backends:
- Gemini Nano via Chrome’s built-in AI APIs
- Transformers.js for running Hugging Face models in-browser
- WebLLM for larger models via WebGPU
Why I built this
I wanted to explore what’s possible with client-side AI in 2025+. As browser capabilities improve (WebGPU, built-in AI APIs), more AI workloads can run locally without sacrificing too much quality. This project is a testbed for those possibilities.
What I owned
- Designed the full client-only app architecture
- Integrated multiple local AI runtimes behind one chat experience
- Built local persistence for conversation history and attachments
- Shipped the project as an installable PWA without backend infrastructure
Technical highlights
- Zero backend — everything runs in the browser
- WebGPU acceleration where available
- Conversation history stored locally
- Model switching between different AI backends
- No servers, no API keys, no data collection
Features
- Local AI Models: Built-in AI (Gemini Nano), Transformers.js, WebLLM
- Complete Privacy: All processing happens in your browser
- Multi-modal: Text, images, documents, audio
- Offline: Works without internet after setup
- PWA: Install like a native app
- File Support: Drag-and-drop attachments
Tech Stack
- React + TypeScript + TailwindCSS
- Vite + PWA
- TanStack Router
- IndexedDB (Dexie)
What this shows
This project shows the kind of AI product engineering I enjoy: understanding new platform capabilities, designing a practical abstraction around messy runtime differences, and shipping a product that users can try directly.
Development
git clone https://github.com/bimsina/browser-chat.gitpnpm installpnpm devModel Types
Built-in AI: Browser native (Chrome/Edge only), instant startup
Transformers.js: ONNX models, good variety, medium size
WebLLM: Large models, most capable, bigger downloads
Browser Support
- Chrome/Edge: Full support
- Firefox: Transformers.js + WebLLM
- Safari: Transformers.js only