A self-hosted multilingual web book reader. Reads EPUB, PDF, CBZ and ZIP, in horizontal, vertical and right-to-left text. Point it at your book folders, scan, and read from any browser — your reading position syncs across devices.
- Shelves — each folder under
/booksbecomes a shelf; create as many as you like. - Scanning — books are imported only when you press Scan. Metadata and covers are read from the files.
- Reader — powered by foliate-js: vertical/RTL text, ruby, two-page spreads, scrolling, adjustable font, margins and themes.
- Progress sync — saved on close, on tab switch, and every 60 seconds.
- Search & filters — width- and kana-insensitive search; filter by status, series, author, tags, rating, format and more.
- Editing — edit metadata per book or in bulk, merge authors and series, extract metadata from file paths with rules.
- AI (optional) — with any OpenAI-compatible endpoint, generate summaries, character lists and relationship maps, or ask questions about chapters (EPUB only).
- Password (optional) — require sign-in per browser.
Images for linux/amd64 and linux/arm64 are published to the GitHub Container Registry on every change to main. No clone needed — save this as compose.yaml in an empty folder:
# compose.yaml
services:
registrum:
image: ghcr.io/registrum-dev/registrum:latest
ports:
- "3000:3000"
environment:
REGISTRUM_PASSWORD: ${REGISTRUM_PASSWORD:-}
# AI features, optional. Any OpenAI-compatible endpoint.
AI_BASE_URL: ${AI_BASE_URL:-}
AI_API_KEY: ${AI_API_KEY:-}
AI_MODEL: ${AI_MODEL:-}
volumes:
- ./data:/data # database and covers
- ./books:/books:ro # your books
restart: unless-stoppedmkdir -p data books
docker compose up -dOpen http://localhost:3000, pick a folder under /books to create a shelf, then press Scan.
Book files are never modified, so /books can be mounted read-only. Database migrations run automatically on startup.
docker compose pull
docker compose up -dlatest follows main. To stay on one build, use a sha-<commit> tag instead (e.g. ghcr.io/registrum-dev/registrum:sha-4cef92c); the available tags are listed on the package page.
The compose.yaml in this repository can also build the image itself:
git clone https://github.com/registrum-dev/Registrum.git
cd Registrum
mkdir -p data books
docker compose up -d --build| Name | Default | Description |
|---|---|---|
REGISTRUM_PASSWORD |
(empty) | Sign-in password. Empty disables sign-in |
PORT |
3000 |
Listening port |
DATA_DIR |
/data |
Database and covers |
BOOKS_DIR |
/books |
Parent folder for shelves |
DATABASE_URL |
file:/data/registrum.db |
SQLite file |
AI_BASE_URL |
(empty) | OpenAI-compatible endpoint, without /chat/completions (e.g. https://openrouter.ai/api/v1). Empty disables AI |
AI_MODEL |
(empty) | Model name as the endpoint spells it (e.g. google/gemini-2.5-pro). Pick one with a long context |
AI_API_KEY |
(empty) | Key for the endpoint. Leave empty for one that wants none, such as Ollama |
CORS_ORIGIN |
(empty) | Only needed when serving the web UI from another origin |
Registrum does not terminate TLS. For access outside your home network, put it behind a reverse proxy such as Caddy or Traefik.
- The AI key is read from
AI_API_KEYand never sent to a browser. Summaries send the full book text to the endpoint (tens of thousands of tokens or more). - Manually edited metadata is never overwritten by rescans. Moved files keep their history if the name and size match.
- Comics and scanned PDFs have no text, so they can't be searched.
- Kanji titles sort by code point, not by reading.
Requires Bun 1.4+.
bun install
mkdir -p data books
bun run db:generate
bun run db:deploy
bun run dev # API on :3000, web on :3001Other scripts: bun run check-types, bun run check (Biome), bun run db:migrate, bun run db:studio.
Stack: Hono + tRPC server (apps/server), React web app (apps/web), shared API logic (packages/api), Prisma + SQLite (packages/db).
MIT. Bundled foliate-js (MIT) includes zip.js (BSD-3-Clause) and pdf.js (Apache-2.0); see apps/web/public/foliate-js/LICENSE.