Documentation that travels where your code travels

What it does Autodoc is an experimental CLI toolkit that crawls a git repository and asks an LLM to write documentation for each file and folder. The resulting docs are stored inside the project itself—so they version-control alongside the code—and can be queried via a natural-language interface that returns specific answers with links back to the relevant source files.

The interesting bit Instead of treating documentation as a separate artifact, Autodoc makes it a native resident of the repo. It also estimates API costs before indexing and, on subsequent runs, only re-indexes files that have changed, which suggests the authors are thinking about operational reality rather than just demo polish.

Key highlights

• Generated docs live in the repository and move with the code through version control.
• Query mode lets developers ask questions about the system and receive answers with direct references to source files.
• Cost estimator previews indexing expenses upfront; incremental re-indexing limits future API spend.
• Per-file model selection automatically picks the cheapest OpenAI model whose context window fits the token count.
• Prompts are partially customizable via configuration, though templates are currently code-centric.

Caveats

• Explicitly experimental and not production-ready; the authors warn that things may break.
• Currently OpenAI-only, with self-hosted models like Llama and Alpaca listed as future work.
• The CLI querying experience is described as “very basic” with room for improvement.
• Files under roughly 4,000 tokens fall back to GPT-3.5, which produces noticeably less accurate documentation; the authors recommend GPT-4 8K minimum.
• Indexing large repositories can cost several hundred dollars in API fees.

Verdict Worth a look if you maintain a complex codebase that newcomers struggle to navigate and you have budget for OpenAI API calls. Skip it if you need battle-hardened tooling or are hoping to run everything offline today.