Agentic Research
Send an agent to research across your docs and code, then write a cited result.
For anything bigger than a quick edit — "document our auth model end to end", "write a rollback section from how deploys actually work" — Docpipe can send a research agent to investigate first, then write. It searches your other documents and your indexed code, works through the problem over several steps, and writes a cited result into your document. It takes a few minutes, and it runs in the background so you can keep working.
Start Research & write
From the Ask AI menu (with or without a selection), choose Research & write. The panel explains what it does — "Researches your codebase and docs, then writes into this document" — and asks you to describe the outcome you want. Type your request into Describe what to write…, then select Start writing.
While it works
A placeholder block appears where the result will land, showing the agent's progress as it goes — searching the codebase, reading related docs, drafting the write-up. You can leave the page; it keeps running. If you change your mind, use the block's Cancel control to stop it.
When it finishes, the researched, cited content replaces the placeholder. If you edited that part of the document while it was working, it appends the result at the end instead of overwriting your changes, and tells you it did.
What it can draw on
The agent researches two sources:
- Your indexed code — it can search and read your connected GitHub repositories. See Code-aware docs for how to index them.
- Your other documents — it searches the rest of your knowledge base and cites what it uses.
Research & write is grounded in what you've actually got — real code and real docs — so the result comes with citations rather than being written from thin air. The more of your code is indexed, the more it has to work from.