Skip to content

Project Wiki

The Project Wiki turns the documents your project already has — design docs, lore, a GDD, architecture notes, READMEs — into a navigable knowledge graph that lives inside the editor. It reads the folder your docs sit in (the “vault”), draws each document as a node on a pan/zoom canvas, and shows the selected note in a reader beside it. Nothing to export, no separate app: it is a tab next to your viewport.

The wiki is a full-panel overlay, reached the same way the Integrations view is — from the panel’s / slash menu → Project Wiki. A back button in the top-left corner returns you to the chat. Opening it on a project with no docs folder yet leaves an empty view and creates nothing.

The vault it reads defaults to your configured agent working directory — the tree your docs live in. The derived index is written under the project’s .blueprintai/local/ area; it is regenerated from the vault, never hand-maintained, and never part of your docs.

Each node is a document. Edges are the links between them, of two kinds:

  • Authored links — a [[wiki-style link]] you wrote in a note. These always show.
  • Glossary auto-links — connections the wiki infers from shared terminology across your notes, without you linking anything by hand. These can be toggled off when you want only the links you authored.

Drag a node to pull the graph around; the layout settles with a force simulation. Pan and zoom the canvas freely. Selecting a node loads it into the reader and recenters the view on it.

The graph draws at one of two grains, toggled from the slash menu:

  • Documents (default) — one node per file.
  • Concepts — each document split into the logical concepts it covers, so a single large design doc becomes several connected concept nodes. Useful for seeing how ideas relate across notes rather than how files do.

The wiki doesn’t stop at prose. It draws your project’s own files and assets — Blueprints, levels, source files — right beside the notes that talk about them: GA_WingedForm next to the ability’s design note, GameplaySkill.h next to the architecture doc that names it. This is on by default; one slash-menu toggle hides it when you want to read only the writing.

A second toggle adds the #include dependencies between those files, turning the map into something you can walk as a codebase. It is off by default — while you are reading the documents, the code is context rather than subject.

Selecting a node opens its body in the reader pane. Below the text, connection chips list the note’s links and backlinks — click one to jump the whole view to that note, highlighting the term that connects them and scrolling to it. Hovering a chip previews where that reference is mentioned in the note you’re on.

A search field at the bottom filters the canvas to matching notes and highlights every match in the reader — including inside fenced code blocks. Alongside it, category chips filter the graph to a topic (design, systems, art, and so on); categories, search, and the toggles all compose, so you can narrow to “systems notes mentioning dash” in a couple of clicks.

There is no manual “re-index” step. When you open the wiki it regenerates the index from the vault, then watches that folder: add, edit, or delete a note and the canvas refreshes on its own. In the editor the work happens on a background process so the UI never stalls; in headless/automation runs it happens inline.

Double-click a node to open what it stands for. An asset opens in its own Unreal asset editor — double-clicking a Blueprint opens the Blueprint editor, because nothing else would do. Anything else opens in whatever application your OS uses for that kind of file: a .docx design doc in Word, a .h in your code editor. The path shown under the reader’s title reveals the file in Explorer / Finder.

The + button creates a new empty markdown page in the vault, re-indexes, and opens it ready to write — so a thought you have while reading the graph becomes a note without leaving the panel.