An Open-Source NotebookLM (for the Audio Overview Part)

”NotebookLM open source” usually means one of two wishes: audit the pipeline that turns text into a two-host conversation, or run it with your own models and storage. This project delivers both — for the audio-overview workflow specifically.

What's actually in the stack

• Script generation: a dialogue prompt (fully editable — it's just a file) that turns an article or book sample into a two-host script with disagreements, questions, and summaries. Works with any OpenAI-compatible LLM, including local models behind a compatible API.
• Synthesis: two distinct neural voices, alternating naturally; swap the TTS provider via one interface.
• Automation: a cron pipeline watches your RSS feeds and book folder — audio appears daily without clicking anything.
• Delivery: standard private podcast RSS + a PWA player. AGPL-3.0, ~1500 lines of TypeScript you can read in an afternoon.

What it deliberately doesn't clone

NotebookLM's interactive Q&A and multi-document research grounding aren't here — this is the listening pipeline, not the research notebook. If your use case is "my daily reading, as audio, automatically," the open-source version is arguably better; if it's "interrogate 40 PDFs," use NotebookLM.

Read the prompt, change the show

Because the dialogue prompt is a plain file, you can change the hosts' personalities, language, or format in one edit — something no closed product lets you do.

FAQ