PRACTICAL GUIDE · DEVELOPER PREVIEW
Use a local AI model in your browser
Nabsun is a free, open-source desktop browser with a bundled Qwen3 1.7B model. You can try the assistant without a cloud account or API key.
Start with a page you know
- Download Nabsun for Windows x64 or an Apple-silicon Mac. Downloads include the model and are about 1.4 GB.
- Open a short public article in Nabsun.
- Click Ask about page. Leave Use this page on in 0.1.13 (called Include page in older versions).
- Ask: “Summarize this page in five bullet points. Separate facts from opinions.”
- Press Send, then compare the answer with the page before relying on it. In 0.1.13 you can also choose an example question; it fills the box and waits for Send.
What works offline?
Local inference runs on your CPU. Once installed, the bundled model does not require internet access to generate a response. Loading websites, downloading models, and using cloud providers still require a connection.
A local model does not make every browser task offline or keep all browsing data on your device: websites still receive normal requests, and the providers and tools you choose determine where task data goes.
Choose tasks that fit the model
Try a short summary, extracting a handful of dates, or rewriting a paragraph. The small bundled model can struggle with long pages and multi-step actions. For more demanding tasks, choose a larger local model through Ollama or connect a supported cloud provider in Settings → Assistant choice in 0.1.13 (Model or Models & accounts in older versions).
Stay in control of actions
Nabsun asks for approval for browser actions by default. Read the proposed action before allowing it. Stop cancels active work and clears pending prompts. The assistant is a developer preview; model answers and site compatibility vary.
Try Nabsun
MIT licensed. Windows x64 and Apple-silicon Mac builds. Cloud providers may require paid accounts. See the security documentation and current limits.