Search page titles, headings and text across the docs.

Browse docs

The Knowledgebase

Browse your project's knowledge as a tree or graph, and see the Organizer that keeps its areas current.

Open Knowledge from a project's header to see its system map. It brings knowledge entries and plans together under areas — parts of your product such as checkout or deployment — rather than leaving you with one long document list.

Four ways to see it

  • Tree is the default. Search areas, knowledge, and plans; expand an area on the left and read its chapter or an entry on the right. Entries and open plans not yet assigned to an area appear under Unfiled.
  • Graph draws areas and documents as connected nodes. Select one to open its reader; pan and zoom to explore the shape, including pinch-to-zoom on a touch screen.
  • Organizer shows the runs that derive the map. You can start a pass with Run now and inspect its progress. It also runs daily when the corpus has changed.
  • History lists the background knowledge-maintenance runs that settle proposals against the code.

An area's reader has separate Knowledge and Plans sections. Plans show open work and shipped history, so you can follow the work alongside the system it concerns. Knowledge entry keys follow their titles; area identities are preserved where the underlying topic still fits.

An area is a chapter, not a folder

A folder holds things. An area explains them. Each one carries:

PartWhat it gives you
NameThe subsystem as your team would say it out loud.
DescriptionWhat the subsystem covers, and where its edges are.
Covered pathsThe repo paths the area maps to, so a claim has an address.
MembersThe knowledge entries and plans filed under it.

Opening an area gives you its chapter, its sub-areas, and the entries and plans filed there — not just a directory listing.

The Organizer rebuilds the map

The Organizer is a built-in automation. A pass summarizes documents, embeds them, rebuilds the areas around their meaning, and curates changed topics. Related entries form leaf areas; related areas can group under parents. A matching area keeps its identity; an empty one can retire. New topics receive names and descriptions, and chapters are written over the resulting tree.

Within a topic, identical entries fold into one. Overlapping or contradicting entries, and entries citing files that no longer exist, are handed to a background maintainer that settles them against the current code and records each run in Knowledge → History. Plans remain separate work items unless they are truly identical. Failed maintenance runs remain on Curate proposals cards with Retry run and Dismiss controls.

Agents navigate the same map

An agent begins with a small manifest of area names and what they cover, not every document body. It can search for the fact it needs by meaning, find every entry that mentions an exact name or path with grep_knowledge, open an area's chapter and index of sub-areas, entries, and plans, then read the relevant document in full. Entries not yet filed under an area remain searchable. Structured memory explains why this keeps context lean.

The same area taxonomy files knowledge, rules, and plans and classifies work threads. That lets you connect the explanation of a system with the plans and threads working on it.

Knowledge lands with the code

What a thread learns is staged with its work: its knowledge updates wait in the Project knowledge section of the side panel's Extras tab, where you can apply or discard them, and land only when the thread merges. That keeps the Knowledgebase from describing code that never shipped. A fact that is already true of main can be saved immediately.