Self-hosted NotebookLM alternative on Railway: Open Notebook

By Tracklayer · Updated 2026-11-20

Short answer: yes. Open Notebook is an open-source, privacy-focused alternative to Google NotebookLM. You collect PDFs, web pages, YouTube videos, audio and notes into notebooks, chat with them with citations, and generate multi-speaker podcasts, using AI providers you choose. Tracklayer’s Railway template runs Open Notebook 1.15.0 with SurrealDB 2.7.0, a password and persistent storage. For code repositories, DeepWiki-Open is a separate template.

Deploy Open Notebook on Railway

When this makes sense

NotebookLM-style tools answer questions from documents you supply, with citations back to the source. Self-hosting suits you if you want:

Be clear about what “private” covers. The sources and notebooks live in your Railway project. The text sent to a hosted model for answers and embeddings goes to that provider.

What the template deploys

Open Notebook’s official image runs three processes: a web app, an API and a background worker that ingests sources, creates embeddings and renders podcasts. They share one data volume for uploads and audio. SurrealDB, which is private to the project, stores notebooks, sources, notes, embeddings, the job queue and encrypted credentials. The template generates every secret, serves the UI and API from one HTTPS domain, and adds health checks so a deploy goes live only after the database is migrated and the app answers.

Deploy Open Notebook, step by step

These steps use Tracklayer’s Open Notebook template.

  1. Deploy. No input is required to start.
  2. Sign in. Enter OPEN_NOTEBOOK_PASSWORD from the Open Notebook service’s Variables tab.
  3. Add a model. Go to Manage, Models, add a credential for your AI provider, test it, and choose default chat and embedding models. Add a text-to-speech model if you want podcasts.
  4. Create a notebook. Add a source and ask questions once processing finishes.
  5. Custom domain. If you add one in Railway, set API_URL and CORS_ORIGINS on the Open Notebook service to https://{your-domain} and redeploy.

Data is persistent: uploads and podcasts are on the app volume and everything else is in SurrealDB. Both images use exact version tags. To upgrade, change the tags and redeploy; migrations run on start. Back up the SurrealDB volume before major upgrades.

What it costs

Railway bills by usage. The template reports a measured 423 MB of RAM for Open Notebook and 212 MB for SurrealDB at idle, and estimates about $7 per month idle and about $23 per month under typical use, which are estimates and not measurements. AI provider usage is billed separately by the provider.

Source processing, embeddings and podcast audio in the worker raise memory use, and the SurrealDB cache grows with data. The template notes that the Trial plan can boot it, but processing large PDFs or podcasts may exceed 1 GB, so Hobby is recommended.

For code repositories: DeepWiki-Open

If your documents are source code, Tracklayer’s DeepWiki-Open template is a different tool. It reads a GitHub, GitLab or Bitbucket repository and writes an interactive wiki with architecture pages, Mermaid diagrams and a chat that answers from the code. It runs as one service behind a password with a volume for clones, embeddings and wikis.

Deploy DeepWiki-Open on Railway

Limits to know

Other options

If you want a general chat interface rather than a document workspace, see the ChatGPT-style UIs on Railway and the guide on a private ChatGPT with Open WebUI and Ollama. For document search with your own vectors, see pgvector for RAG. All guides are at /guides/.

FAQ

How do I log in to Open Notebook?

Open the service URL and enter OPEN_NOTEBOOK_PASSWORD from the Open Notebook service's Variables tab. The same value works as a Bearer token for the REST API. Keep OPEN_NOTEBOOK_ENCRYPTION_KEY unchanged, because it encrypts the API keys stored in the app.

Do I need an AI provider account?

Yes. After login, go to Manage, Models, add a credential for a provider such as OpenAI, Anthropic, Google or OpenRouter, and choose default chat and embedding models. Podcasts also need a text-to-speech model. Provider usage is billed by the provider, not by Railway.

Does my data stay private?

Notebooks, sources and embeddings are stored in SurrealDB inside your Railway project. Text you send to a hosted AI provider for chat or embeddings goes to that provider. Only a model you run yourself, for example Ollama on another server, keeps prompts off third-party services.

What are the licences?

Open Notebook is MIT. SurrealDB is under the Business Source License 1.1, which converts to Apache-2.0 after its change date; self-hosting for your own application is permitted and offering SurrealDB as a commercial database service is not. DeepWiki-Open is MIT.

Can it run more than one replica?

No. The Open Notebook web app, API and worker share one data volume, so it runs as one replica. Add RAM and CPU for large documents or parallel podcast jobs.