The self-hosted NotebookLM alternative everyone is starring
Open Notebook just crossed 25,000 GitHub stars. It's MIT-licensed, 100% local, and supports 18+ AI providers including OpenAI, Anthropic, Ollama, and LM Studio. Drop in PDFs, YouTube links, audio, and web pages — get answers with citations. Generate multi-speaker podcasts with 1-4 voices. Your research data never leaves your machine.
Two-minute Docker install
# Grab the compose file
curl -o docker-compose.yml \
https://raw.githubusercontent.com/lfnovo/open-notebook/main/docker-compose.yml
# Change OPEN_NOTEBOOK_ENCRYPTION_KEY to a long random string (32+ chars)
# Launch
docker compose up -d
docker compose ps
open http://localhost:8502
Ports: 8502 UI, 5055 REST API, 8000 SurrealDB (internal). Persistent data lives in ./surreal_data and ./notebook_data.
Three setups that actually work
- Cheap & fast — Anthropic Claude Haiku. Best for citation-heavy long-context research.
-
Free + local — Ollama with llama3 or qwen2.5. Point Open Notebook at
http://host.docker.internal:11434. - Best podcasts — OpenAI gpt-4o for the script, TTS-1-HD for multi-speaker audio. Configure 1-4 speaker profiles in Settings → Podcasts.
Real use-cases
- Research → cited paper. 10-20 PDFs + a few web pages. Open-ended questions return cited answers. Beats ChatGPT for serious lit review.
- Content production. Transcripts plus competitor blog posts in, outlines + multi-speaker podcast episodes out.
- Legal review. Run it 100% on-prem with Ollama. NDAs, MSAs, policies — no client data ever crosses the network.
60-second walkthrough
I broke down the install + the dashboard in a short here: https://www.youtube.com/shorts/6WN52Mw3n28
The full audio breakdown lives on the GPTAIClips podcast on Spotify: https://open.spotify.com/show/033sVcOc9PNwH6eIX5SPSq
More short-form breakdowns of trending AI repos at https://gptaiclips.com.
Affiliate disclosure: I use GoHighLevel (Pro trial via that link) as the CRM I sell AI workflows out of — referral link, doesn't change your price.
Top comments (0)