Organised the way readers look
Architecture, endpoints, integrations and schemas grouped by domain, not one noisy page per file.
Open source · MIT · Not backed by YC
DeepDoc reads your repository and writes a docs site with a chatbot that cites real files, then regenerates only the pages your commits touch.
A real run on Pranav322/youtube-notes · 6m 12s, shown in 42 seconds
Music: “Aitech” by Kevin MacLeod (incompetech.com), licensed under CC BY 3.0.
What it does
Architecture, endpoints, integrations and schemas grouped by domain, not one noisy page per file.
An assistant ships with the site. Every answer points to a file path and line range in your source.
After a commit, only the affected pages regenerate.
Colours, logo and navigation re-apply from the saved plan. No LLM calls.
Pipeline
Every page is grounded in scanned evidence and checked against the source before the site builds. Only two phases spend tokens.
Parses your source locally and maps what is there: routes, models, background jobs, integrations and OpenAPI specs.
No LLMClusters the call graph into domains, then asks a model to name them and arrange them into sections.
A few callsAssembles the evidence for each page, writes it, validates it against the source and repairs what fails.
One call per pageIf the repository has an OpenAPI spec, it is staged as an interactive reference alongside the written pages.
No LLMWrites the site itself: navigation, search, theme and the optional chatbot. Ready to preview or deploy.
No LLMGet started
Sign in with GitHub, pick a repository (private ones too) and get a docs site at a stable address. Private by default.
Open DeepDoc CloudInstall the CLI, bring your own model, and keep everything on your machine. Works with a local model too.
Read the quickstartWorks with
Endpoint, runtime and call-graph analysis cover these stacks. Other languages may parse, but are not fully supported.