Open Notebook: Private, Self-Hosted Gemini Notebook (NotebookLM) Alternative

Open Notebook is an open-source, self-hosted AI research and note-taking web app. It is a free, private local Gemini Notebook (formerly NotebookLM) alternative that runs through Docker and works with cloud providers or local models from Ollama and LM Studio.

Your notebooks can hold PDFs, Office files, EPUB books, web pages, Reddit posts, YouTube videos, audio, and pasted text. Open Notebook extracts and indexes the material for search, cited chat, AI-generated notes, custom transformations, and podcasts with up to four speakers.

The app, database, and uploaded sources run on your computer or server. A fully local model setup keeps AI processing on that infrastructure. Cloud model providers receive the context sent to their APIs. You control the provider, model assignment, source context, storage, and deployment.

Features

• Multi-Model AI Support: Connect more than 18 provider integrations for language models, embeddings, speech-to-text, and text-to-speech: OpenAI, Anthropic, Google, Vertex AI, Azure OpenAI, Groq, Mistral, DeepSeek, xAI, OpenRouter, DashScope, MiniMax, Ollama, LM Studio, and other OpenAI-compatible endpoints. Each task may have its own model assignment.
• Source Processing: PDFs, Word documents, PowerPoint slides, Excel files, EPUB books, Markdown, text, HTML, audio, video, YouTube links, web pages, Reddit links, and more.
• Configurable Extraction Engines: Content Core handles ingestion. Optional Docling adds OCR and layout-aware document parsing. Optional Crawl4AI renders JavaScript-heavy pages locally or connects to a remote Crawl4AI server. Firecrawl, Jina, and Simple URL extraction are also available.
• Multi-Speaker Podcasts: Turn notebook material into audio with one to four speakers. Episode profiles define the topic, length, tone, structure, and model choices. Speaker profiles store personas and voices. A separate voice model may be assigned to each speaker.
• Granular Context Control: Set each source to Full Content, Summary, or Excluded for a chat. Shorter context lowers token use with cloud APIs.
• Content Transformations: Apply built-in or custom prompts to summarize documents, extract key points, identify methods, compare themes, or create structured notes. Assign a specific model to any transformation that needs it.
• Search and Cited Chat: Full-text and vector search cover the material stored in your notebooks. References in Chat and Ask responses open the relevant source, note, or insight.
• Notebook Organization: Keep sources, notes, chat sessions, and generated audio grouped by project. The interface also provides list and tile views, recent notebooks, sortable source columns, and a command palette.
• REST API and MCP Access: The REST API exposes notebooks, sources, notes, search, chat, models, and settings. The separate Open Notebook MCP server connects those capabilities to compatible clients such as Claude Desktop and VS Code extensions.
• Credential Management: Add provider credentials in the web interface, test connections, discover available models, and assign defaults. Open Notebook encrypts stored credentials with the deployment encryption key.
• Reasoning and Math Display: Reasoning-capable models such as DeepSeek and Qwen show collapsible thinking sections. KaTeX renders inline and display mathematics in chat and other content views.
• Optional Password Protection: A shared application password protects the web interface and API. It provides basic access control for a single-user or small shared deployment, not enterprise identity management.

Use Cases

• Academic Research: Upload papers, book chapters, lecture recordings, and data tables. Search across them, compare methods, save cited answers as notes, and create an audio review for material that needs another pass.
• Content Research: Collect articles, interview audio, statistics, and source documents in one notebook. Transformations extract claims, quotes, themes, and outline material while the original sources stay available for checking.
• Technical Documentation: Store API documentation, architecture notes, issue discussions, and specifications on your own server. The REST API and MCP server also make the collection available to development assistants.
• Learning and Study: Build a notebook around one course or skill, add mixed media, ask source-grounded questions, and generate summaries or custom study notes. Multiple chat sessions separate topics inside the same project.
• Confidential Research: Local models, embeddings, and speech services keep document processing inside your infrastructure. Internet-facing deployments need HTTPS, a strong password, restricted CORS origins, backups, and normal server hardening.

Open Notebook vs Gemini Notebook (NotebookLM)

Open Notebook and Google’s NotebookLM both turn a collection of sources into a workspace for research and learning. Open Notebook focuses on self-hosting, model choice, automation, and customizable audio. NotebookLM offers a managed service with Google Drive integration, mobile access, and more ready-made Studio outputs.

Basic Usage

1. Install Docker Desktop on Windows, macOS, or Linux. The recommended deployment uses Docker Compose with separate Open Notebook and SurrealDB services.

2. Create an empty folder and download the current Compose file from the official repository:

curl -o docker-compose.yml https://raw.githubusercontent.com/lfnovo/open-notebook/main/docker-compose.yml

3. Open docker-compose.yml and replace change-me-to-a-secret-string in OPEN_NOTEBOOK_ENCRYPTION_KEY. Keep this key with your backups. Stored provider credentials become unreadable if you lose or change it.

4. Start the services from that folder:

docker compose up -d

5. Wait for the containers to start, then open http://localhost:8502. Port 8502 serves the web interface, port 5055 serves the REST API, and the current Compose file binds SurrealDB’s port 8000 to localhost.

6. Open Models, add a provider configuration, test the connection, sync the models you want to register, and assign defaults. Chat and embeddings require model assignments. Podcast generation also needs suitable language and text-to-speech models.

7. Create a notebook and add sources. Processing runs in the background. A completed source becomes searchable and available for chat, transformations, notes, and podcast generation.

8. Select the context for each chat. Use Full Content for close analysis, Summary for background material, and Excluded for sources that do not belong in the current conversation.

Running Open Notebook Locally with Ollama

Here’s an Ollama Compose example for local language and embedding models. Download that example, pull at least one model, register the Ollama credential in Open Notebook, then assign language and embedding defaults. LM Studio works through the OpenAI-Compatible provider and normally uses http://host.docker.internal:1234/v1 from Docker.

A local deployment works offline after the application images and model files are present. URL imports, cloud model APIs, and provider model discovery need network access. Local inference speed depends on the selected model, system memory, CPU, and optional GPU.

Pros

• Your data stays on your infrastructure.
• Eighteen-plus AI providers are supported.
• Local Ollama models run offline.
• Podcasts support one to four speakers.
• MIT licensing permits commercial modification.

Cons

• Docker setup requires technical confidence.
• You maintain updates, backups, and security.
• Cloud models create separate API costs.
• Local models need capable hardware.

Alternatives and Related Resources

• Gemini Notebook: AI Research Assistant for Students and Professionals

FAQs

Q: How does Open Notebook differ from ChatGPT file uploads?
A: Open Notebook keeps sources, notes, searches, and chat sessions organized inside persistent notebooks. You can reuse the same research collection across conversations, control the context level of each source, and connect the collection to other applications through the REST API or MCP server.

Q: What happens to my data when I move servers?
A: Back up the notebook_data and surreal_data directories together with the encryption key. Restore them on the new host before starting the Compose services. The encryption key is required to read stored provider credentials.

Q: How should I protect an internet-facing deployment?
A: Set OPEN_NOTEBOOK_PASSWORD, put the service behind HTTPS, restrict CORS_ORIGINS to the exact frontend domains, and block direct public access to the application and database ports. Open Notebook uses one shared password. Deployments that need user accounts, roles, or audit logs require additional access controls.

Q: Why does a generated podcast sound robotic?
A: Voice quality depends on the configured text-to-speech model, speaker profile, and generated script. Try another registered voice model, adjust the speaker profile, and review the transcript before generating the audio again. Local speech models often trade natural delivery for privacy and lower usage costs.

Q: What computer does Open Notebook need?
A: The documented minimum is 4GB of RAM and 2GB of application storage, plus room for documents. Eight gigabytes of RAM and more storage are recommended. Local AI models and Docling require additional memory, disk space, and optional GPU capacity.

Q: Why is Docling or Crawl4AI unavailable in Settings?
A: Both local runtimes are optional. Add OPEN_NOTEBOOK_ENABLE_DOCLING=true or OPEN_NOTEBOOK_ENABLE_CRAWL4AI=true to the container environment and restart it. The first start downloads the selected runtime to the data volume.

Last Updated: July 17, 2026