TabMind - Local Tab Saver (Built-in AI)

TabMind is a Chrome extension that turns the pages you are reading into rich, searchable notes. It runs fully on-device using Chrome's built-in AI capabilities to produce TL;DR summaries, key points, and smart tags that stay in your local IndexedDB.

Privacy: 100% local. No servers, API keys, or third-party calls. After Chrome downloads the Gemini Nano models one time, everything works offline.

Features

• One-click Save current tab from the popup or the side panel. Captures page text, current selection, and metadata before handing everything to the on-device models.
• Instant library access in the side panel with editable cards, ratings, tag management, and inline search filters.
• Natural language recall powered by local re-ranking, letting you type prompts like "show me the CSS grid article" and have the best matches rise to the top.
• Markdown export of the current result set from the side panel so you can keep a snapshot of links, ratings, tags, and summaries.
• Clean local storage via IndexedDB; data never leaves the browser.

Install (Unpacked)

1. Open chrome://extensions and enable Developer mode.
2. Click Load unpacked and select the folder TabMind_Local_Tab_Saver.
3. Pin TabMind from the toolbar and click it to open the popup.
4. Use Initialize AI once so Chrome can download the on-device models (requires a user gesture).
   • Track progress in the UI or at chrome://on-device-internals.

Using the extension

• Save as you browse: press Save current tab in the popup or in the side panel to capture the tab immediately. The extension enriches the note in the background with summaries, tags, and entities.
• Search and filter: the side panel supports field filters (title, tags, description) and ordering (date or rating). Natural language queries automatically trigger re-ranking when multiple matches exist.
• Edit and rate: click the pencil icon on any saved card to adjust tags, TL;DR, or key points. Star ratings help surface your best references.
• Export results: click the export icon next to the side panel title to download the current search results as a Markdown file. The export includes the title, rating, tags, summary, and key points for each entry.

How it stays local

• Uses Summarizer API (Summarizer.create() / Summarizer.summarize) for TL;DR and key points.
• Uses Prompt API (LanguageModel.create() / session.prompt) with a JSON Schema constraint to extract tags, intent, and entities, and to power semantic re-ranking.
• Both APIs are built into Chrome and execute Gemini Nano locally once the models are downloaded.

Known limitations (future polish ideas)

• Built-in AI APIs are not yet available in service workers, so all AI calls happen in the popup or side panel contexts.
• Very large libraries may benefit from adding an OPFS/IndexedDB full-text index and chunked re-ranking.
• The initial model download needs a user gesture and enough disk space; if the download fails Chrome will report it in chrome://on-device-internals.

Visual guides

Docs

• Prompt API (Gemini Nano in-browser): https://developer.chrome.com/docs/ai/prompt-api
• Summarizer API (Chrome 138+): https://developer.chrome.com/docs/ai/summarizer-api
• Extensions + AI overview: https://developer.chrome.com/docs/extensions/ai