GradMark-AI

Local AI bookmark organizer powered by Parallax. Organize thousands of bookmarks in 30 seconds with zero cloud, zero leaks.

Features

• Import Chrome bookmarks via Netscape HTML export
• AI-powered organization with two modes:
  • Fix mode: Keep existing folders, only move misplaced bookmarks
  • Rebuild mode: Creates a fresh two-level hierarchy (top folder + subfolder) by batching bookmarks in sets of five so the LLM never exceeds its context window. The CLI streams token throughput (tok/s) in real time while each batch runs.
• Natural language search
• Export to Chrome-compatible HTML without touching the original file
• Statistics with AI insights

Prerequisites

• Node.js 18+
• Parallax running locally on port 3001
• Use a 7B-or-larger model for quality; tiny models like Qwen-0.6B degrade folder suggestions.
• Always keep a backup of your original bookmark export before running organize to avoid data loss.

See the Parallax Quick Start Guide for installation instructions.

Installation

npx gradmark-ai <command>

Or install globally:

npm install -g gradmark-ai

Usage

Import bookmarks

Export from Chrome (Bookmarks Manager → ⋮ → Export Bookmarks) and point the CLI at the HTML file:

npx gradmark-ai import ~/Downloads/bookmarks.html

Organize bookmarks

Start interactive organization:

npx gradmark-ai organize

You'll be prompted to:

1. Choose organization mode (fix or rebuild)
2. Watch streaming throughput (tokens/sec) as each 5-bookmark batch is processed
3. Review changes
4. Optionally export immediately

Search bookmarks

Search using natural language:

npx gradmark-ai search "React state management"
npx gradmark-ai search "machine learning tutorials"

Export bookmarks

Write a fresh Chrome-compatible HTML file (the original export stays untouched):

npx gradmark-ai export --format html

View statistics

Get insights about your bookmark collection:

npx gradmark-ai stats

How it works

1. Bookmarks are stored locally at ~/.gradmark-ai/bookmarks.json
2. AI processing uses Parallax's local LLM (default: Llama-3-8B) with streaming responses so you can see live throughput
3. No data leaves your machine
4. Batch processing splits requests into 5-bookmark chunks to stay within local context limits during rebuild mode

Configuration

The tool connects to Parallax at http://localhost:3001 by default. Make sure Parallax is running before using organization features:

parallax run

To use a different port, either set GRADMARK_PARALLAX_PORT or pass --parallax-port <port> to any command:

GRADMARK_PARALLAX_PORT=3030 npx gradmark-ai organize
# or
npx gradmark-ai --parallax-port 3030 organize

For setup instructions, see the Parallax Quick Start Guide.

License

MIT

Support & Feedback

Questions or ideas? Reach out via bojack's GitHub profile. I'm happy to hear how GradMark-AI works for you and what you'd like to see next.