Open Notebook
lfnovo · AI assistants
GitHub stars
40,919Awaiting first sync
No measured star history yet
What is Open Notebook?
Open Notebook gives researchers and teams a NotebookLM-style workspace that runs on their own server. You decide which AI provider sees your material, or keep everything local with models run through Ollama. Instead of a monthly subscription, you pay only for the AI usage you choose.
Open Notebook is a self-hosted, open-source alternative to Google's NotebookLM. You collect PDFs, web pages, videos, audio and Office files into notebooks, then search them, chat with them using cited answers and turn them into podcast episodes. It works with more than 20 AI providers, including OpenAI, Anthropic, Google and local models through Ollama.
Core capabilities
- Organise PDFs, web pages, YouTube videos, audio and Office files into separate notebooks.
- Chat with your research and get answers that cite the sources and notes they used.
- Generate podcast episodes with one to four speakers from your sources and notes.
Notebooks, sources and search
Sources can be files, web links or pasted text, and one source can belong to several notebooks. Supported files include PDF, Word, PowerPoint, Excel, EPUB, Markdown, MP3, WAV, MP4 and MOV, and YouTube links bring in the video transcript. Each source is processed in the background, and Open Notebook offers full-text and vector search across everything you add. You can write notes yourself or have AI generate them.
- Notebooks
- Vector search
- YouTube
- Notes
Chat, transformations and podcasts
Chat answers draw on your research and cite the sources, notes and insights they used, and you control which content is shared with the model. Transformations are saved prompts that run on a source and store the output as an insight, with six built in, including Paper Analysis and Dense Summary. Podcasts are made in three steps, outline, transcript and audio, using speaker and episode profiles you configure. A REST API and an MCP integration let other tools, such as Claude Desktop and VS Code, work with your notebooks.
- Citations
- Transformations
- Podcasts
- REST API
- MCP
Run it with Docker Compose
Open Notebook is licensed under MIT. The official docker-compose.yml runs two services: SurrealDB as the database, and the Open Notebook container with the web interface, API and background worker, which opens on port 8502. You set an encryption key that protects stored API keys, then add a provider and its models on the Models page. There is no password by default, so the README advises binding the ports to localhost or setting OPEN_NOTEBOOK_PASSWORD on a shared network.
- MIT
- Docker Compose
- SurrealDB
- Self-hosted
Where it fits
Use cases
- 01
Build a private research library for a project
Create a notebook and open the Add Source wizard to add URLs, upload files or paste text. Tick Enable embedding for search so vector search and Ask can find the content. Audio and video files are transcribed with the Speech-to-Text model you set under Manage, then Models.
- 02
Summarise every new document the same way
Go to Manage, then Transformations, and create a saved prompt such as a key-points summary, or use built-in ones like Dense Summary and Key Insights. Tick it in the Process step when adding a source, and the result is saved as an insight on that source. Try new prompts first in the Playground tab.
- 03
Turn research into an audio briefing
Set up a speaker profile with a text-to-speech model and one to four speakers, then an episode profile with outline and transcript models, a number of segments and a briefing. Pick the sources and notes to include and start an episode. Generation runs in the background worker, so you can close the browser while it works.