Quick Verdict
Hoarder is the better choice if you save a lot of content and want AI to automatically tag and categorize it. Shiori is the better choice if you want a minimal, lightweight bookmark manager that archives full pages and runs on almost any hardware. They represent opposite ends of the bookmark manager spectrum — smart automation vs. bare-bones efficiency.
Overview
Hoarder is a relatively new bookmark manager that uses AI (via local models or OpenAI) to automatically tag and organize saved links. It focuses on the "save now, organize later" workflow — you dump links and let the AI figure out the categorization.
Shiori is a Go-based bookmark manager inspired by the old Pocket experience. It's a single binary with embedded SQLite — no external dependencies, no AI, no frills. You save pages, it archives the content for offline reading, and you tag things manually.
Feature Comparison
| Feature | Hoarder | Shiori |
|---|---|---|
| Language | TypeScript (Next.js) | Go |
| License | AGPL-3.0 | MIT |
| Database | SQLite or PostgreSQL | SQLite, PostgreSQL, or MySQL |
| AI tagging | Yes (local or OpenAI) | No |
| Auto-categorization | Yes (AI-powered) | No |
| Page archiving | Screenshots + text extraction | Full readable content |
| Browser extensions | Chrome, Firefox | Chrome (beta), Firefox |
| Mobile app | PWA + iOS/Android (beta) | PWA |
| CLI | No | Yes |
| API | Yes | Yes |
| Full-text search | Yes | Yes |
| Pocket import | Yes | Yes (Netscape format) |
| Multi-user | Yes | Limited |
| Docker image | ~500 MB+ | ~30 MB |
| Minimum RAM | ~512 MB | ~30-50 MB |
Installation Complexity
Hoarder requires more infrastructure. The Docker Compose includes the main app, a Chromium container for page screenshots, and optionally a local AI model (Ollama). The stack uses 500+ MB of RAM at minimum, and more with local AI inference.
Shiori runs as a single container with an embedded SQLite database. No external services needed. Total footprint is ~30-50 MB of RAM. You can run it on a Raspberry Pi Zero without issues.
Winner: Shiori. Dramatically simpler setup and smaller footprint.
Performance and Resource Usage
| Metric | Hoarder | Shiori |
|---|---|---|
| RAM (idle) | ~300-500 MB (with Chromium) | ~30-50 MB |
| RAM (with AI) | ~2-4 GB (with local Ollama) | N/A |
| CPU | Moderate (AI inference, rendering) | Very low |
| Disk | Screenshots + DB | Archived text + DB |
| Startup time | 10-30 seconds | <2 seconds |
Hoarder is 10-15x heavier on RAM due to the Chromium rendering engine and optional AI backend. If you're running on limited hardware or alongside many other services, this matters.
Winner: Shiori. Runs on almost any hardware.
Content Archiving
Both save page content, but differently:
- Hoarder takes screenshots (visual snapshots) and extracts text. The screenshots provide a visual reference but can't be searched. Text extraction enables full-text search.
- Shiori downloads and parses the full page into readable content (similar to Reader View in Firefox). The result is searchable text with formatting preserved, available for offline reading.
Winner: Shiori for offline reading and searchability. Hoarder for visual reference (screenshots show the exact page layout).
Organization
This is where they diverge most:
- Hoarder's AI tagging analyzes saved content and applies tags automatically. You configure your preferred AI backend (OpenAI or local Ollama) and the AI processes each bookmark. The more you save, the more useful the automatic organization becomes.
- Shiori's manual tagging requires you to add tags yourself when saving or editing bookmarks. It's predictable and doesn't require AI infrastructure.
Winner: Hoarder if you save dozens of links daily and don't want to categorize each one. Shiori if you prefer manual control and don't want AI dependencies.
Community and Support
| Metric | Hoarder | Shiori |
|---|---|---|
| GitHub stars | ~4,000 | ~11,000+ |
| First release | 2024 | 2019 |
| Development pace | Very active | Steady (less frequent releases) |
| Contributors | Growing | 67 |
Shiori has a longer track record and larger community. Hoarder is newer but developing rapidly.
Use Cases
Choose Hoarder If...
- You save large volumes of content and hate manual tagging
- You already run Ollama or have OpenAI API access
- You want screenshot previews of saved pages
- You prefer a modern Next.js UI
- You have 512 MB+ RAM available for bookmark management
Choose Shiori If...
- You want the lightest possible bookmark manager
- You run on limited hardware (Raspberry Pi, small VPS)
- You prefer offline-first archiving (full readable content)
- You want a CLI for scriptable bookmark management
- You don't need AI features
- You value a battle-tested, mature project
Final Verdict
For most users saving bookmarks casually (10-20 per week), Shiori is the better choice — it does the job with minimal resources and no complexity. The full content archiving is genuinely useful for offline reading.
For power users who save 50+ links weekly and need help organizing them, Hoarder justifies its higher resource cost with AI tagging that saves real time. If you already run an AI stack, the marginal cost of adding Hoarder is minimal.
Frequently Asked Questions
Does Hoarder require a GPU for AI tagging?
No. Hoarder's AI tagging works with cloud APIs (OpenAI, Ollama) or local CPU-based models. A GPU accelerates local model inference but is not required. With Ollama on CPU, tagging takes a few seconds per bookmark instead of sub-second with a GPU.
Can Shiori archive pages for offline reading?
Yes. Shiori downloads and stores the full HTML content of saved pages. You can read archived pages even if the original site goes offline. The archive is stored locally in Shiori's database, not as separate files.
Does Hoarder have a browser extension?
Yes. Hoarder has browser extensions for Chrome and Firefox, plus a mobile app for iOS and Android. Shiori also has browser extensions but no official mobile app — you access it through the mobile browser.
How much storage does each tool use?
Shiori uses minimal storage — typically 10-50 MB for hundreds of bookmarks since it stores compressed HTML. Hoarder uses more storage due to AI embeddings and richer metadata — expect 100-500 MB for a similar collection, more if you enable screenshot archiving.
Can I import bookmarks from Pocket or Raindrop.io?
Both support importing bookmarks from HTML files (the standard browser bookmark export format). Hoarder also supports direct Pocket import. Neither has a built-in Raindrop.io importer, but exporting from Raindrop as HTML works.
Which is better for a team or family?
Hoarder supports multiple users with shared lists — better for collaborative use. Shiori is designed for single-user use. For shared bookmark management, also consider Linkwarden which has full team support with shared collections and permissions.
Top comments (0)