← All articles
black hardbound books on shelves

Karakeep (Hoarder): Self-Hosted AI-Powered Bookmark Manager

Productivity 2026-02-14 · 6 min read karakeep hoarder bookmarks read-later ai ollama
By Selfhosted Guides Editorial Team — Self-hosting practitioners covering open source software, home lab infrastructure, and data sovereignty.

We all have the same problem: you find something interesting online, bookmark it, and never see it again. Browser bookmarks become a graveyard of unsorted links. Read-later apps like Pocket help, but they are SaaS services that own your data and charge subscription fees. Karakeep (formerly known as Hoarder) takes a different approach: it is a self-hosted bookmark manager that uses AI to automatically tag and categorize everything you save. Links, notes, images -- throw anything at it and the AI organizes it for you.

Photo by Jake Lorefice on Unsplash

Karakeep bookmark manager dashboard showing saved links with AI-generated tags

What Makes Karakeep Different

Most bookmark managers require you to manually organize your saves into folders or apply tags. This is why they fail -- nobody maintains their bookmark taxonomy over time. Karakeep flips this by using a local AI model (via Ollama or OpenAI) to automatically generate relevant tags when you save something.

Save an article about Docker networking? Karakeep reads the page content and tags it with "docker," "networking," "containers," "devops" automatically. Save a recipe? It gets tagged with "cooking," "recipe," and the cuisine type. You never have to think about organization.

Core features:

Self-Hosting with Docker

Karakeep needs three services: the app itself, a Meilisearch instance for full-text search, and a Chrome browser for page screenshots and content extraction. Optionally, add Ollama for local AI tagging.

Create your compose file:

version: "3.8"

services:
  karakeep:
    image: ghcr.io/hoarder-app/hoarder:latest
    container_name: karakeep
    restart: unless-stopped
    depends_on:
      - meilisearch
      - chrome
    ports:
      - "3000:3000"
    volumes:
      - karakeep-data:/data
    environment:
      MEILI_ADDR: "http://meilisearch:7700"
      BROWSER_WEB_URL: "http://chrome:9222"
      DATA_DIR: "/data"
      NEXTAUTH_SECRET: "your-secret-at-least-32-characters"
      NEXTAUTH_URL: "https://bookmarks.yourdomain.com"
      # For local AI with Ollama (recommended)
      OLLAMA_BASE_URL: "http://host.docker.internal:11434"
      INFERENCE_TEXT_MODEL: "llama3.2"
      INFERENCE_IMAGE_MODEL: "llava"
      # Or use OpenAI instead
      # OPENAI_API_KEY: "sk-your-key"

  chrome:
    image: gcr.io/zenika-hub/alpine-chrome:latest
    container_name: karakeep-chrome
    restart: unless-stopped
    command:
      - --no-sandbox
      - --disable-gpu
      - --disable-dev-shm-usage
      - --remote-debugging-address=0.0.0.0
      - --remote-debugging-port=9222
      - --hide-scrollbars

  meilisearch:
    image: getmeili/meilisearch:v1.12
    container_name: karakeep-meilisearch
    restart: unless-stopped
    volumes:
      - meilisearch-data:/meili_data
    environment:
      MEILI_NO_ANALYTICS: "true"

volumes:
  karakeep-data:
  meilisearch-data:

Start everything:

docker compose up -d

Navigate to http://your-server:3000 and create your account. The first user becomes the admin.

Setting Up AI Tagging with Ollama

The AI tagging is what sets Karakeep apart, and running it locally with Ollama means no API costs and no data leaving your server.

If you do not already have Ollama running, install it on your host:

curl -fsSL https://ollama.com/install.sh | sh

Pull the models Karakeep uses:

# Text model for analyzing page content and generating tags
ollama pull llama3.2

# Vision model for analyzing images (optional)
ollama pull llava

The text model (llama3.2 at ~2GB) handles most tagging. The vision model (llava at ~4.7GB) is optional but enables AI analysis of saved images and screenshots. If your server has limited RAM, skip the vision model.

How the AI Tagging Works

When you save a link:

  1. Chrome fetches the page and takes a screenshot
  2. Meilisearch indexes the full text content
  3. The text content is sent to your configured LLM
  4. The LLM analyzes the content and returns relevant tags
  5. Tags are attached to the bookmark automatically

Processing happens asynchronously, so saving is instant. Tags appear within a few seconds to a minute depending on your hardware and model size.

Tuning Tag Quality

The quality of auto-generated tags depends on your model choice:

You can also manually add or remove tags after the AI generates them. Over time, the combination of AI suggestions and your manual refinements creates a well-organized collection.

Like what you're reading? Subscribe to Self-Hosted Weekly — free weekly guides in your inbox.

Browser Extensions and Mobile Apps

Browser Extension

Install the Karakeep extension for Chrome or Firefox. Configure it with your server URL and API key (generated in Settings > API Keys). Then:

The extension sends the URL to your server, which handles fetching, screenshot, indexing, and AI tagging.

Mobile Apps

Karakeep has native iOS and Android apps. Configure them with your server URL and credentials. The mobile apps support the share sheet, so you can share links from any app directly to Karakeep.

This is particularly useful for saving content from social media apps, Reddit, YouTube, and anywhere else you discover things on your phone.

Full-Text Search

Meilisearch powers Karakeep's search, and it is excellent. Search queries match against:

Search is typo-tolerant and fast. Even with thousands of bookmarks, results appear instantly.

Managing Your Collection

Beyond AI tags, Karakeep provides several organizational tools:

The search and AI tagging together mean you rarely need manual organization. Save everything, search when you need something specific.

Comparison with Alternatives

Karakeep vs Linkwarden

Linkwarden is a polished bookmark manager with collections, tags, and page archiving. It does not have AI tagging -- organization is entirely manual. If you are disciplined about organizing bookmarks, Linkwarden's clean interface and collaboration features are excellent. If you want "save and forget" with AI handling the organization, Karakeep is the better choice.

Karakeep vs Wallabag

Wallabag is a mature read-later application focused on article reading. It strips pages down to readable content, supports offline reading, and has good mobile apps. Wallabag is optimized for reading articles; Karakeep is optimized for collecting and finding anything. Use Wallabag if your primary workflow is "save articles to read later." Use Karakeep if you want a general-purpose collection manager.

Karakeep vs Raindrop.io

Raindrop.io is a SaaS bookmark manager with a beautiful UI, nested collections, and full-text search on Pro ($3/month). It has no AI features but has the most polished interface in the category. Karakeep matches its search capabilities, adds AI tagging, and keeps everything self-hosted. Raindrop.io wins on UI polish; Karakeep wins on privacy and AI features.

Karakeep vs Shiori

Shiori is a minimalist bookmark manager inspired by Pocket. It is lightweight and simple -- save links, read them later. No AI, limited search, basic tagging. If you want maximum simplicity with minimal resource usage, Shiori works. Karakeep is for people who want intelligence applied to their bookmark collection.

Resource Usage

Karakeep's stack uses moderate resources due to the Chrome and Meilisearch components:

Total: ~1-2 GB without Ollama active, 5-10 GB during AI processing. Ollama unloads models from memory after inactivity, so the steady-state usage is manageable.

A machine with 8 GB RAM can run Karakeep with a small Ollama model comfortably. For larger models or heavy usage, 16 GB is recommended.

Backup

Back up the data volume and Meilisearch data:

# Stop services to ensure consistency
docker compose stop

# Back up volumes
docker run --rm -v karakeep-data:/data -v $(pwd):/backup alpine \
  tar czf /backup/karakeep-data-$(date +%Y%m%d).tar.gz /data

docker run --rm -v meilisearch-data:/data -v $(pwd):/backup alpine \
  tar czf /backup/meilisearch-data-$(date +%Y%m%d).tar.gz /data

# Restart
docker compose start

When to Use Karakeep

Karakeep is ideal for digital hoarders -- people who save lots of content and struggle to organize it. If you have hundreds of browser bookmarks across devices, a Pocket account full of unread articles, and screenshots of things you wanted to remember, Karakeep gives you a single place to dump everything with AI handling the filing. Save first, search later.

Get free weekly tips in your inbox. Subscribe to Self-Hosted Weekly