NEURON

Local AI · No Cloud · Fully Private

Initializing runtime…

🤏 Pick a smaller model — Qwen 0.5B is 3–4× faster than 1.5B for simple questions.
📐 Reduce context window — drop from 4K to 2K if you don't need long conversations.
🌡️ Lower temperature — values near 0 generate tokens faster with less exploration.
🗂️ Close other tabs — WebGPU competes for GPU memory with other open pages.
🌐 Wired connection for first load — faster download = faster initial model cache.
🖥️ Use Chrome or Edge on desktop — best WebGPU driver support for inference speed.

Welcome to NEURON

A fully private, client-side AI assistant powered by WebLLM and WebAssembly. Your data never leaves your device.

1Select a model from the sidebar and click Load Model

2Optionally upload documents to enable RAG

3Start chatting — everything runs locally!

Model runs entirely in your browser · No data sent anywhere