BROWSER-LOCAL AI
Qwen 2.5 0.5B · Lightweight
On-device inference
↓Prepare your local AIChecking browser compatibility…⌄

Download once, then disconnect. A compact model runs directly on your device’s GPU.

Several hundred MB download~1.1 GB GPU memory*WebGPU required
*Runtime estimate; actual memory and speed vary. Small models can be inaccurate.
✳

A little intelligence.
Entirely yours.

Think something through. Rewrite a sentence.
Start a conversation that stays on your device.

Download a model to begin. Internet is only needed for the initial setup.
◇ Your words stay here

Local means local.

Prompts and replies are processed by WebLLM on this device. No online AI API, analytics, or chat-upload endpoint is used by this app.

The first model download contacts Hugging Face and GitHub’s model-library host. Those hosts receive normal connection information, not your chat. The website host also receives ordinary page requests.

App files and model files are cached in this browser. Use “Check offline model” after setup to reload and generate a short test with model network access blocked. Offline reopening also depends on the app cache staying intact.

Browser storage can be cleared or evicted. Private browsing may not retain downloads. Chats are stored locally, not encrypted by this app; anyone with access to this browser profile may read them. Export a chat before clearing it.

The small Qwen model has limited reasoning, no live news, and can make mistakes. Older conversation turns are omitted when context grows.

WebLLM · Qwen 2.5 model & license