GleanMark

Your personal knowledge base with semantic search and AI integration.

GleanMark lets you save web articles as bookmarks with their content, then find them later by meaning — not just keywords. Ask "what did I save about leadership challenges" and get the right articles, even if those exact words never appeared. Everything runs locally on your machine, your data stays yours, and AI agents can tap into your knowledge base directly.

Why GleanMark

The problem: You read dozens of articles every week. Some are worth revisiting — for research, writing, or reference. But when you need them months later, keyword search fails because you don't remember the exact words. And your bookmarks are just a graveyard of URLs with no context.

What GleanMark does differently:

• Saves content, not just URLs — When you bookmark an article, GleanMark captures the text so you can search against it later, even if the original page goes offline
• Finds by meaning — Hybrid search combines semantic understanding (vector embeddings) with keyword matching (BM25), so searching "team motivation strategies" finds your Harvard Business Review article about employee morale
• Works with AI agents — Expose your knowledge base to Claude via MCP. Ask "what articles did I save about X?" and your AI assistant searches your personal library directly
• Local-first — Runs entirely on your machine by default: no accounts, no API keys, embeddings on-device, data stored locally. An optional hosted Cloud (invite-only) adds cross-device sync — see below
• Grows more valuable over time — As you accumulate hundreds of bookmarks, GleanMark becomes a personal corpus that reflects your interests and research, ready to be queried, referenced, and cited

Features

• Hybrid Search — Semantic (vector) + keyword (BM25) combined via Reciprocal Rank Fusion
• Multilingual — MultilingualE5 embeddings + jieba Chinese word segmentation
• Desktop App — Tauri-based app with system tray, global hotkey (Cmd+Shift+G), and auto-updater
• Web UI — Search, browse, and manage bookmarks at http://localhost:21580
• Browser Extension — Chrome extension (Manifest V3) for one-click bookmark saving
• CLI — Terminal interface for power users
• MCP Server — AI agent integration for Claude Code / Claude Desktop
• Dark Mode — Light / dark / system theme support
• Local Embeddings — fastembed-rs runs on-device, no API keys needed
• Import / Export — JSON-based data portability

Quick Start

Desktop App (Recommended)

Download the latest release from GitHub Releases:

• macOS: .dmg (Apple Silicon)
• Windows: .msi installer
• Linux: .deb or .AppImage

macOS: If you see a security warning, run xattr -cr /Applications/GleanMark.app (the app is not yet notarized by Apple).

The app auto-manages Qdrant and embedding models. First launch may take a moment to download models (~100MB, one-time only).

CLI

cargo install --path crates/gleanmark-cli

# Add a bookmark
gleanmark add https://example.com/article --title "Great Article" --tags "rust,programming"

# Search
gleanmark search "async programming patterns"

# List all bookmarks
gleanmark list

# Start the HTTP server + Web UI
gleanmark serve

Browser Extension

Install from the Chrome Web Store, then:

1. Start the server: gleanmark serve (or use the Desktop App)
2. Click the GleanMark icon on any page to save it

Developers: You can also load the extension/ folder as an unpacked extension via chrome://extensions/.

Cloud (optional, invite-only)

By default everything is local. If you'd rather sync bookmarks across devices without running Qdrant locally, GleanMark offers a hosted Cloud backend: embeddings are still computed on your device, and only the pre-computed vectors + content are sent to your private, per-account collection.

Cloud is invite-only for now. Open Settings → "Interested in GleanMark Cloud?" in the app to request access (and tell us whether you'd pay for it). Once you have an account, switch Settings → Backend → Cloud and sign in — no URLs or keys to configure.

MCP Server (AI Agents)

cargo build -p gleanmark-mcp --release

Claude Code (.claude/settings.json):

{
  "mcpServers": {
    "gleanmark": {
      "command": "/path/to/gleanmark-mcp"
    }
  }
}

Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "gleanmark": {
      "command": "/path/to/gleanmark-mcp"
    }
  }
}

Architecture

gleanmark/
├── gleanmark-core       Core library: embeddings, storage, search
├── gleanmark-server     HTTP API + Web UI (Axum, port 21580)
├── gleanmark-cli        CLI tool (clap)
├── gleanmark-tauri      Desktop app (Tauri v2)
├── gleanmark-mcp        MCP server for AI agents (rmcp)
└── extension/           Chrome extension (Manifest V3)

All components share gleanmark-core. Data stored in ~/Library/Application Support/gleanmark/ (macOS) or ~/.local/share/gleanmark/ (Linux).

API Reference

The HTTP server runs on http://127.0.0.1:21580.

Method	Endpoint	Description
GET	/api/health	Health check
POST	/api/bookmarks	Create bookmark
GET	/api/bookmarks?limit=50&offset=<id>	List bookmarks (paginated)
DELETE	/api/bookmarks/{id}	Delete bookmark
POST	/api/search	Search bookmarks
POST	/api/export	Export to JSON file
POST	/api/import	Import from JSON file
POST	/api/open	Open URL in system browser

Search Request

{
  "query": "async programming",
  "limit": 10,
  "tags": ["rust"]
}

Search Response

[
  {
    "bookmark": {
      "id": "uuid",
      "url": "https://...",
      "title": "...",
      "content": "...",
      "tags": ["rust"],
      "created_at": "2026-04-02T...",
      "updated_at": "2026-04-02T..."
    },
    "score": 0.85
  }
]

MCP Tools

Tool	Description	Key Parameters
search_bookmarks	Semantic + keyword search	query, limit? (default 10), tags?
list_bookmarks	List all bookmarks	limit? (default 20), offset?
add_bookmark	Save a new bookmark	url, title, content?, tags?
delete_bookmark	Delete by ID	id

Development

Prerequisites

• Rust 1.85+ (edition 2024)
• Qdrant binary in PATH or ~/.local/share/gleanmark/bin/
• For desktop app: Tauri v2 prerequisites

Build

git clone https://github.com/jiayun/gleanmark.git
cd gleanmark

# Build all
cargo build

# Build specific crate
cargo build -p gleanmark-server

# Desktop app (dev mode)
cd crates/gleanmark-tauri
cargo tauri dev

# Run tests
cargo test

# Re-index bookmarks (after model changes)
cargo run -p gleanmark-cli -- reindex

Project Structure

crates/
  gleanmark-core/
    src/
      embedding.rs      MultilingualE5 + SPLADE via fastembed
      storage.rs        Qdrant operations (CRUD, scroll pagination)
      search.rs         Hybrid search (RRF fusion)
      qdrant_manager.rs Qdrant process lifecycle
      models.rs         Data types and config
  gleanmark-server/
    src/main.rs         Axum routes + rust-embed static serving
    static/             Web UI (HTML/CSS/JS)
  gleanmark-cli/
    src/main.rs         CLI commands via clap
  gleanmark-tauri/
    src/
      main.rs           Tauri setup, Axum background task, splash screen
      tray.rs           System tray + global hotkey
  gleanmark-mcp/
    src/
      main.rs           MCP server entry (stdio transport)
      server.rs         Tool handlers (lazy-init GleanMark)
      types.rs          MCP tool parameter schemas
extension/
  manifest.json         Chrome Manifest V3
  popup/                Bookmark save UI
  background/           Service worker

Tech Stack

Component	Technology
Vector DB	Qdrant (embedded)
Dense Embeddings	MultilingualE5-Small (384d, fastembed)
Sparse Embeddings	SPLADE PP v1 + jieba segmentation
Web Framework	Axum
CLI	clap
Desktop	Tauri v2
MCP	rmcp
Browser Extension	Chrome Manifest V3

License

MIT OR Apache-2.0