Browser AI chat

Try a small local language model without installing a desktop app. You choose when the download starts.

Try Qwen3 0.6B on your device

This small model is useful for trying simple local chat. It can make mistakes and is not a substitute for a stronger model on demanding tasks. This tool uses a 4,096-token context and reserves up to 512 tokens for each reply.

Loading downloads model files from Hugging Face and WebLLM’s model library from GitHub. The download is several hundred megabytes and uses browser storage. WebLLM estimates roughly 1.4–1.9 GB of GPU memory, depending on supported precision; actual device requirements can be higher.

Messages stay in memory in this tab and are not uploaded. Reloading closes the conversation. Downloaded files can remain on this device for the next visit; the browser may remove cached files when storage is low.

No model has been downloaded by this page.

Temporary conversation

After loading, try: “Explain what a token is in two sentences.”

For larger models or a browser without WebGPU, choose a local app for your computer. There is no cloud fallback.

What you need

Use an up-to-date browser and graphics driver with WebGPU support, enough free memory, and a connection for the initial model download. Browser and device support varies. The tool checks support before starting the download and reports failures rather than sending your messages elsewhere.

How to try local chat

Choose Download and load model, then wait for the ready message.
Ask a short question and watch the answer appear.
Use Stop generation if you have enough. Clear conversation starts fresh without deleting model files.
Use Delete downloaded model files to reclaim the storage used by this tool.

This is a small demonstration model with a fixed 4,096-token context. Long conversations and documents can exceed that limit. It does not search the internet or read files. Compare larger local models and check your memory when you need more capability.

Sources

Qwen3 0.6B model card and Apache 2.0 license; WebLLM; WebGPU browser support.