# Deep research

> Two agents prepare a verified document from code or company documents: the analyst writes, the reviewer checks every claim against the sources.

**“Deep research”** is a mode in which two agents, the **analyst** and the **reviewer**, work through a question in depth **using code or company documents** — connected [knowledge sources](/en/docs/connector/knowledge), trackers and [your own MCP servers](/en/docs/connector/custom-mcp), read-only — and prepare a document.

- You ask a question; the agents build a **living document** that grows as they go. An example about code: “where does the delay in the export come from”; an example about documents: “Compare the three versions of the commercial proposal for Romashka LLC by price, discount and term” — a document with a table and a link to each version.
- There is a **conversation** for refining the direction; the base commit is shown.
- The finished document can be exported to **PDF** and **Markdown**. If the research was requested with the `@buff research …` command from a tracker or a document, the answer goes there too: in Notion as a child page, in Google Drive as a new document in the same folder, in other systems as a comment with a PDF or a link.
- In a development project, connected sources, trackers and your own MCP servers are read on a par with the code — if the connection names this project and is running.
- In a [space without code](/en/docs/quickstart/space) the analyst and the reviewer work from the team’s documents and the knowledge base rather than code: they compare sources, name contradictions, and every claim links to a page or entry. Research takes no attachments — attach the file to a question in “Ask” or put it into a connected source. When a question is enough and when you need a research — [Question or research?](/en/docs/ask-vs-research).