Local Nano

Tracking since May 28, 2026.

Changelog

• Jun 4, 2026
  description

  Local Nano is a private AI assistant that lives one keypress away on every web page — and it runs the language model entirely inside your browser. No account, no API key, no servers, no data collection.
  
    Press Ctrl+Shift+K (Cmd+Shift+K on Mac) to open the panel. You can:
  
    • Ask about the page you're on — your question is answered using the page's visible text as context.
    • Rewrite selected text in place — highlight a sentence, tell Local Nano how to change it, and the rewrite streams directly back into the page. One click Undo restores the original.
  
    How it stays private
    All inference runs locally via WebGPU (with an automatic CPU fallback) using Transformers.js and ONNX Runtime Web. The only network request Local Nano ever makes is a one-time download of the open model weights from Hugging Face, which are then cached on your device. Your page content, your prompts, and your chat history never leave your machine — there is no telemetry and nothing is sent to a remote server.
  
    What to expect
    The first run downloads the model (a few GB) and takes roughly 30–90 seconds, with a progress indicator; after that it loads from cache. A modern GPU with a few GB of memory is recommended. If the model can't load at full quality, Local Nano automatically tries lighter precision and CPU modes, and if your device can't run it at all, it tells you clearly instead of failing silently, with a copyable diagnostic you can share in a bug report (nothing in it leaves your device automatically).
  
    Open and on-device by design — for anyone who wants a quick AI helper on the pages they're reading without handing their browsing or their text to a cloud service.

  Local Nano is a private AI assistant that lives one click away on every page. The language model runs entirely inside your browser. No account, no API key, no servers, no data collection, and no remote code. Everything that runs ships inside the package.
  
  How to open it
  
  Click the Local Nano icon next to the address bar.
  
  Or press Ctrl+Shift+K (Cmd+Shift+K on Mac). On a fresh install the shortcut may need to be bound once at chrome://extensions/shortcuts. The icon's tooltip shows the current binding, and the icon works without one.
  
  What you can do
  
      - Ask about the page you're on. Your question is answered using the page's visible text as context. The chat is DOM aware.
  
      - Rewrite selected text in place. Highlight a sentence, tell Local Nano how to change it, and the rewrite streams directly back into the page. One click Undo restores the original.
  
      - Choose your model. A gear popover offers a curated catalog of on-device models. Pick one, click Load to switch. Your choice persists across sessions.
  
      - Reclaim memory when you walk away. After a configurable idle period (5, 15, 60 minutes, or Never) the model is released from VRAM. The next use re-warms automatically.
  
  How it stays private
  
  All inference runs locally via WebGPU, with an automatic CPU/WASM fallback. The runtime is Transformers.js and ONNX Runtime Web, both bundled inside the extension. The only network request Local Nano ever makes is the one-time download of the open model weights from Hugging Face, which are cached on your device afterward. Your page content, prompts, and chat history never leave your machine. No telemetry. No analytics. No remote scripts.
  
  What to expect
  
  The first run downloads the model (a few GB) with a real-time percentage indicator and takes roughly 30 to 90 seconds. After that it loads from cache. A modern GPU with a few GB of memory is recommended. If the model can't load at full quality, Local Nano automatically tries lighter precision and CPU modes. If your device can't run it at all, it tells you clearly and gives you a copyable diagnostic. Nothing in the diagnostic leaves your device automatically.
  
  Open and on-device. For anyone who wants a quick AI helper on the pages they're reading without handing their browsing or their text to a cloud service.

• Jun 4, 2026
  short_description

  Local Nano is a private AI assistant that lives one keypress away on every web page — and it runs the language model entirely…

  Local Nano is a private AI assistant that lives one click away on every page. The language model runs entirely inside your browser.…

• Jun 4, 2026
  permissions

  storage, offscreen

  storage, offscreen, alarms

Permissions & access

Permissions

storageoffscreenalarms

Host access

https://huggingface.co/*, https://*.huggingface.co/*, https://cdn-lfs.huggingface.co/*

Screenshots