Deploy and Host Open Notebook on Railway

Open Notebook is an open-source, privacy-first alternative to Google's NotebookLM. Bring your own AI provider keys, upload sources, chat with your documents, generate notes and podcasts, and keep every notebook on infrastructure you control.

About Hosting Open Notebook

This template runs Open Notebook as two services: the application — a single container that runs the FastAPI backend, the background task worker and the Next.js web UI together — and a dedicated SurrealDB database that stores your notebooks, sources, notes and embeddings on a persistent Railway volume. The web UI proxies API calls to the backend internally, so one public domain serves the whole app; there is no separate reverse proxy to configure.

Two things this template gets right that a bare deploy does not:

Both secrets are set. OPEN_NOTEBOOK_ENCRYPTION_KEY encrypts the AI provider keys you save in the database, and OPEN_NOTEBOOK_PASSWORD turns on password authentication so your notebooks — and the API keys the app drives — are not exposed to anyone who finds the URL.
Every image is pinned. The app is pinned to lfnovo/open_notebook:1.14.0 and the database to surrealdb/surrealdb:v2.6.5 — the SurrealDB major (v2) that Open Notebook is built and tested against. No floating tags means a redeploy never silently swaps the engine or the database version under your data.

Common Use Cases

Research assistant over your own documents: upload PDFs, web pages and transcripts, then ask questions grounded in your sources.
Notebook-to-podcast: turn a set of sources into a generated audio overview, NotebookLM-style, using your own TTS provider.
Private, multi-provider AI workspace: connect OpenAI, Anthropic, Google, Groq, Ollama or others and keep the whole workspace self-hosted.

Dependencies for Open Notebook Hosting

Deployment Dependencies

Open Notebook — the upstream application (lfnovo/open-notebook).
SurrealDB v2 — the database backing every notebook, source and embedding.

Implementation Details

First login: open the app URL and enter the value of the OPEN_NOTEBOOK_PASSWORD variable (find it under the service's Variables tab in Railway). Then open Settings → API Keys and add at least one AI provider key — Open Notebook ships with no model configured, so this first step is required before you can create sources or chat.
Your data lives in SurrealDB: notebooks, sources, notes and vector embeddings are all stored in the database volume, so they survive restarts, redeploys and version changes. The app's own volume holds uploads and processing caches.
Optional heavy extractors: set OPEN_NOTEBOOK_ENABLE_DOCLING=true or OPEN_NOTEBOOK_ENABLE_CRAWL4AI=true to enable OCR / advanced document parsing and local JavaScript-rendered web scraping. These download large ML models and a browser on first boot (hundreds of MB to a few GB, cached on the volume afterwards) and raise memory use noticeably — leave them off unless you need them.
Resource usage: measured on a fresh deploy, the two services idle at roughly 0.5 GB of RAM combined (app ~0.35 GB, SurrealDB ~0.2 GB). Memory grows with ingestion, embedding and generation work, so expect a few dollars a month when idle and to scale toward ~$10–25/month under regular use — more if you enable the Docling or Crawl4AI runtimes or process large media files.

Why Deploy Open Notebook on Railway?

Railway provisions the SurrealDB database, both persistent volumes and the public domain as part of one deploy, so Open Notebook comes up configured, authenticated and reachable on the first try — no compose file, reverse proxy or manual database wiring. The encryption key and access password are generated for you, and the pinned images keep every redeploy reproducible.