repo-story
What if you could listen to a codebase like a podcast?
repo-story surveys a repository, researches its ideas, and produces a long-form documentary narration — then speaks it aloud with AI voice synthesis.
How it works
The pipeline runs in five phases:
- Survey — scan the repo structure, read key files, understand the architecture
- Explore — dive deep into interesting patterns, trace data flows, map dependencies
- Synthesize — distill findings into a narrative outline with chapter structure
- Research — connect repo concepts to broader engineering ideas and prior art
- Narrate — write the full script and render it to audio via Chatterbox TTS
The output is a multi-chapter M4B audiobook, typically 3-5 hours long, with chapter markers for navigation.
What makes it interesting
Most code documentation describes what code does. repo-story tells the story of why it exists — the design philosophy, the engineering tradeoffs, the ideas the authors were wrestling with. It's closer to a documentary than a README.
The audio on the home page is a five-hour deep dive into the engineering behind gstack and autoresearch.