LazyBookmarks

LazyBookmarks is a Chrome extension that uses Google's in-browser Gemini Nano model to automatically organize your bookmarks. It's a testbed and proof-of-concept for experimenting with Chrome's Prompt API — the built-in, on-device language model that runs entirely in your browser with no cloud calls. Bookmark things lazily without worrying about folders, then click "Organize" and let the AI classify everything into your existing folder structure.

lazybookmarks-demo-web.webm

Quick Start

git clone https://github.com/LLMCoolJ/lazybookmarks.git

1. Open chrome://extensions
2. Enable Developer mode (top-right toggle)
3. Click Load unpacked → select the lazybookmarks folder
4. Click the extension icon → Organize

Requires Chrome 138+ with Prompt API flags enabled. See full installation guide below.

This is experimental software. It modifies your bookmarks by moving them between folders. Back up your bookmarks before using it.

Back Up Your Bookmarks

Before installing, export your bookmarks:

1. Open Chrome and go to chrome://bookmarks
2. Click the three-dot menu (top-right of the Bookmark Manager)
3. Select Export bookmarks
4. Save the HTML file somewhere safe

You can restore from this file at any time using Import bookmarks from the same menu.

Installation

Requirements

Google Chrome 138+ only. The Prompt API is Chrome-specific — Chromium, Edge, Brave, and other variants are not supported.

Hardware minimums for Gemini Nano (source):

• GPU: >4 GB VRAM — or CPU: 16 GB RAM, 4+ cores
• Storage: 22 GB free on the volume containing your Chrome profile
• OS: Windows 10/11, macOS 13+, Linux, or ChromeOS (Chromebook Plus)

Step 1: Enable Chrome Flags

Set both of these to Enabled, then relaunch Chrome:

1. chrome://flags/#optimization-guide-on-device-model
2. chrome://flags/#prompt-api-for-gemini-nano-multimodal-input

Step 2: Verify the Model

Go to chrome://on-device-internals and confirm Gemini Nano is downloaded. If it hasn't started downloading, the flags may need a second relaunch. See Chrome's troubleshooting guide if the model doesn't appear.

Step 3: Load the Extension

git clone https://github.com/LLMCoolJ/lazybookmarks.git

1. Open chrome://extensions and enable Developer mode
2. Click Load unpacked → select the lazybookmarks folder
3. Pin the extension, click the icon, and hit Organize

Project Overview

lazybookmarks/
├── manifest.json        Extension manifest (Manifest V3)
├── background.js        Service worker — detects new bookmarks, manages the pending queue
├── popup/               Extension popup — opens the viewer tab
├── viewer/              Main bookmark viewer UI — tree view, search, organize button
├── ai/                  AI classification engine — Prompt API, structured prompts, suggestions
├── lib/                 Shared utilities — bookmark store, tree rendering, search
├── debug/               Live AI debug log viewer — prompts, responses, streaming events
├── icons/               Extension icons (16, 48, 128px)
├── vendor/              Third-party libraries (Orama full-text search)
└── docs/                Cached Chrome API documentation for LLM-assisted development

Detailed File Reference

ai/ — AI Classification Engine

The core of the extension. Orchestrates a multi-phase pipeline: build a taxonomy of your existing folders, cluster uncategorized bookmarks by theme, then classify each one into the best-fit folder.

File	Purpose
ai-organizer.js	Orchestrates the full classification pipeline (taxonomy, clustering, classification)
language-model.js	Wrapper around Chrome's LanguageModel (Prompt API) — session creation, streaming, quota management
prompts.js	All prompt templates and JSON schemas used for structured output
suggestions-ui.js	UI for reviewing AI suggestions — accept, reject, or apply all moves
taxonomy-cache.js	Caches the folder taxonomy so it can be reused across classification batches
undo.js	Snapshot-based undo/redo for bookmark moves
debug-log.js	Structured event logging for the debug viewer
debug-constants.js	Debug configuration flags

lib/ — Shared Utilities

File	Purpose
bookmark-store.js	chrome.bookmarks API wrapper — loads the full tree, normalizes it into flat and hierarchical formats
tree-view.js	DOM tree renderer with expand/collapse, drag targets, and folder highlighting
search.js	Full-text bookmark search powered by Orama
search-view.js	Connects the search input to the search engine and renders results
favicon.js	Fetches favicons for bookmark URLs

viewer/ — Main UI

File	Purpose
viewer.html	Main page with HTML templates for folders and bookmarks
viewer.js	Orchestrator — wires together the tree, search, AI organizer, and suggestions UI
viewer.css	Styling with dark/light theme support

debug/ — Debug Viewer

A standalone page (debug/debug.html) that displays a live stream of AI classification events — every prompt sent to Gemini Nano, every streamed response token, parse results, and errors. Open it from the extension's debug page or navigate to it directly.

docs/ — Cached Documentation

Pre-fetched documentation from developer.chrome.com, docs.orama.com, and other sources. These are used for LLM-assisted development — providing context to AI coding tools without needing live web access. Not required at runtime.