Hey everyone (-_-) I'm Aman, I built n0x. I use Claude Code for actual coding work but for everything else.. "explain this concept" "summarize this PDF" "search the web and tell me what matters"- I was bouncing between ChatGPT, Grok, and Gemini. None of those tasks need a 70B model behind an API. They just need a decent model and a clean interface. So I built something that runs entirely in the browser. You open a tab, pick a model (40+ open-source ones, from 360MB to 70B), and it downloads once then loads from cache and after that it works offline. What's under the hood: LLM inference runs on your GPU via WebGPU in a Web Worker (UI stays smooth at 60fps) Document Q&A uses hybrid search - keyword matching + vector search combined not just cosine similarity There's an agent mode that chains tools (search, docs, Python, memory) and you watch it think step by step Image generation, voice input/output, conversation branching, persistent memory The privacy part is real.. your prompts, files, and model weights never leave the browser. The only network calls happen if YOU use web search or image gen. If WebGPU isn't enough, you can plug in Ollama or any OpenAI-compatible API (Groq, OpenRouter) and switch mid-conversation. Same tools, bigger models. Try it, break it, tell me what sucks. I use this daily and I want to make it better.

N0X

Run LLMs, search, and RAG in your browser. No server.

Run LLMs, search, and RAG in your browser. No server.