Bookmark Hub

A self-hosted bookmark management application that helps you organize, search, and discover your saved links. Built with Rust, featuring AI-powered summarization and tagging to make your bookmarks more useful and discoverable.

Features

• Offline-First: Store and manage bookmarks entirely on your own infrastructure
• AI-Powered Organization: Automatic tagging and summarization with multi-provider LLM support (Ollama, OpenAI, Anthropic, Gemini, OpenRouter)
• RAG-Enhanced Search: Intelligent search using Retrieval-Augmented Generation to find relevant bookmarks based on semantic similarity
• Full-Text Search: Search through bookmark titles, URLs, content, and AI-generated summaries
• Tag Management: Organize bookmarks with manual and AI-suggested tags
• Content Extraction: Automatically extract and store readable content from web pages
• Modern Web Interface: Responsive WebAssembly-based frontend built with Yew
• REST API: Complete API for programmatic access and integrations
• CLI Tools: Command-line interface for batch operations and automation

How to Run

Quick Start (Docker)

The easiest way to get started is using Docker Compose:

# Build the application container
$ just build-container

# Start the full stack (database + application)
$ docker compose up

This will start:

• PostgreSQL database on port 5432
• Browserless Chrome automation service on port 3001
• Bookmark Hub server on port 3000
• Web interface accessible at http://localhost:3000

Using with Host Ollama

If you already have Ollama running on your host machine, use the alternative compose file:

# Build the application container
$ just build-container

# Start with host Ollama integration
$ docker compose -f docker-compose.host-ollama.yml up

This configuration:

• Uses network_mode: host for direct access to host services
• Connects to Ollama running at localhost:11434
• Uses pgvector/pgvector:pg17 for vector embedding storage
• Uses default Ollama models for AI features

Development Setup

For local development with hot reloading:

Prerequisites

• Rust toolchain with wasm32-unknown-unknown target
• Just task runner
• Trunk for WebAssembly builds
• PostgreSQL database
• Optional: Ollama for local AI features, or an API key for a cloud provider (OpenAI, Anthropic, Gemini, OpenRouter)

Running Components

1. Start PostgreSQL database (locally or via Docker)

2. Start the server:

   $ just run-server

   This starts the API server with development configuration.

3. Start the frontend (in a separate terminal):

   $ just run-spa

   This starts the development server at http://localhost:8080 with hot reloading.

Configuration

The server accepts configuration via environment variables:

# Database
export PG_HOST=localhost
export PG_PORT=5432
export PG_USER=your_user
export PG_PASSWORD=your_password
export PG_DATABASE=bookmark_hub

# Application
export HMAC_KEY=your_secret_key
export APP_DATA_DIR=/path/to/data

# Optional: Chrome Automation
export CHROME_HOST=localhost
export CHROME_PORT=3001

LLM Provider Configuration

AI features (tagging, summarization, embeddings, RAG) are disabled when LLM_TEXT_MODEL is not set. To enable them, configure a provider:

Ollama (local, default):

export LLM_PROVIDER=ollama
export OLLAMA_URL=http://localhost:11434
export LLM_TEXT_MODEL=qwen3.5:4b
export LLM_EMBEDDING_MODEL=qwen3-embedding:0.6b
export LLM_EMBEDDING_DIMENSION=1024

OpenAI:

export LLM_PROVIDER=openai
export OPENAI_API_KEY=sk-...
export LLM_TEXT_MODEL=gpt-4o
export LLM_EMBEDDING_MODEL=text-embedding-3-small
export LLM_EMBEDDING_DIMENSION=1536

Anthropic (requires a separate embedding provider since Anthropic has no embedding API):

export LLM_PROVIDER=anthropic
export ANTHROPIC_API_KEY=sk-ant-...
export LLM_TEXT_MODEL=claude-sonnet-4-20250514
export LLM_EMBEDDING_PROVIDER=openai
export LLM_EMBEDDING_API_KEY=sk-...
export LLM_EMBEDDING_MODEL=text-embedding-3-small
export LLM_EMBEDDING_DIMENSION=1536

Gemini:

export LLM_PROVIDER=gemini
export GEMINI_API_KEY=AIza...
export LLM_TEXT_MODEL=gemini-2.5-flash
export LLM_EMBEDDING_MODEL=gemini-embedding-001
export LLM_EMBEDDING_DIMENSION=768

OpenRouter:

export LLM_PROVIDER=openrouter
export OPENROUTER_API_KEY=sk-or-...
export LLM_TEXT_MODEL=anthropic/claude-sonnet-4
export LLM_EMBEDDING_PROVIDER=openai
export LLM_EMBEDDING_API_KEY=sk-...
export LLM_EMBEDDING_MODEL=text-embedding-3-small
export LLM_EMBEDDING_DIMENSION=1536

You can mix providers — for example, use Anthropic for text and OpenAI for embeddings via LLM_EMBEDDING_PROVIDER and LLM_EMBEDDING_API_KEY.

Variable	Default	Description
LLM_PROVIDER	ollama	Text completion provider
LLM_TEXT_MODEL	(none, disables AI)	Model for text tasks
LLM_EMBEDDING_PROVIDER	(falls back to LLM_PROVIDER)	Embedding provider
LLM_EMBEDDING_MODEL	(falls back to LLM_TEXT_MODEL)	Model for embeddings
LLM_EMBEDDING_DIMENSION	1024	Embedding vector dimension
LLM_EMBEDDING_API_KEY	(none)	API key for embedding provider if different
LLM_REQUEST_TIMEOUT_SECS	120	HTTP request timeout for LLM calls
OLLAMA_URL	http://localhost:11434	Ollama base URL
OPENAI_API_KEY	(none)	OpenAI API key
ANTHROPIC_API_KEY	(none)	Anthropic API key
GEMINI_API_KEY	(none)	Google Gemini API key
OPENROUTER_API_KEY	(none)	OpenRouter API key

CLI Usage

# Login (required first)
$ just run-cli login --url http://localhost:3000 --username your_user --password your_password

# Add a single bookmark
$ just run-cli add --url https://example.com

# Add multiple bookmarks from a file (one URL per line)
$ just run-cli add-batch --file urls.txt

Testing

Run end-to-end tests using Hurl (requires running application):

$ hurl --verbose --test test.hurl

Architecture

• Server: Axum-based REST API with background processing daemons
• Frontend: Yew WebAssembly application for modern web experience
• Database: PostgreSQL for reliable data storage with vector embeddings
• AI Integration: Multi-provider LLM support via rig-core (Ollama, OpenAI, Anthropic, Gemini, OpenRouter)
• RAG System: Vector similarity search using embeddings for intelligent bookmark discovery
• Content Processing: dom_smoothie for web page content extraction
• Browser Automation: Browserless Chrome for reliable web page rendering and content extraction