NotebookLM's Audio Overviews proved that two AI hosts discussing your documents is a genuinely good way to absorb information. But it's manual, closed, and lives inside Google. Here's the self-hosted, always-on version of that idea.
With NotebookLM you upload a document, click, wait, and download. Repeat for every document, every day. listenwhilerunning inverts this: you declare your sources once — RSS feeds, subreddits, a folder of ebooks — and a pipeline generates audio on schedule, publishing everything to one private podcast feed your podcast app already follows. New audio just appears.
| listenwhilerunning | NotebookLM Audio Overview | |
|---|---|---|
| Workflow | Declare sources once; a daily pipeline generates audio automatically | Upload a document and click Generate, per document |
| Delivery | Standard podcast RSS — Apple Podcasts, lock-screen controls, resume everywhere | In-app playback (plus download) |
| Whole books | Epub/PDF becomes a serialized audiobook with an AI hosts' intro episode | A single overview of the document |
| Interactive Q&A | No — you can't interrupt the hosts | Yes, interactive mode with questions |
| Multi-document research | No — it's a listening pipeline, not a research notebook | Yes, deep grounding across a source set |
| Openness | AGPL-3.0 source; your storage, your API keys, any OpenAI-compatible LLM | Closed; lives inside your Google account |
| Account required | None — no login, no vendor in the loop | Google account |
| Cost | Cloudflare free tier + pennies of LLM usage (~$0/mo) | Free tier with limits; higher limits via Google AI subscription tiers |
Fairness matters: NotebookLM's interactive mode (interrupting the hosts to ask questions) and its deep multi-document grounding are ahead. If you need conversational Q&A over a research corpus, use NotebookLM. If you want your daily information diet turned into audio automatically, that's what this project is for.
The format is similar — two hosts, natural back-and-forth, disagreements and summaries. Voice quality depends on the TTS you configure; the default neural voices are close to human narration.
Yes. Any OpenAI-compatible endpoint works (OpenRouter, a local model behind a compatible API, etc.). Without an LLM it degrades to plain readout.
A hosted version is under evaluation — join the waitlist on the homepage and you'll be notified at launch.