Skip to Content

Features

AI Wiki

The AI Wiki is a markdown knowledge base that agents build up about your codebases as they work. Instead of rediscovering the same architectural facts on every task, agents write down what they learn, so every task starts smarter than the last.

The wiki lives on disk as plain markdown at ~/.modula/<workspace>/wiki/, with Obsidian-style [[wiki-links]] between pages. Read and edit it in the app’s AI Wiki page, or open the folder in any markdown editor.

How it grows

Agents with the AI Wiki skill enabled treat the wiki as their shared memory:

  • The Researcher reads relevant pages before investigating a task, cites them in its specs, and writes new findings back.
  • The Worker reads pages before implementing, and corrects them when the code contradicts what’s written.
  • The Code-Reviewer reviews against the wiki and records new patterns it surfaces.

What belongs in the wiki is durable codebase truth: architectural patterns and the reasoning behind them, module responsibilities, conventions, quirks, and cross-project contracts. What doesn’t: task-specific decisions, review chatter, logs, or anything your project’s own CLAUDE.md already says. The wiki complements your repos, it never restates them.

Structure

wiki/
├── SCHEMA.md     The conventions agents follow when reading and writing
├── index.md      Catalog of every page, one line each
├── log.md        Chronological log of every wiki change
├── general/      Workspace-wide pages that span projects
└── <project>/    One folder per project

index.md is the entry point agents scan first, and log.md is an append-only history of what was added, updated, or contradicted, and when. When new findings contradict an old page, agents mark the relationship in frontmatter rather than silently rewriting history.

Editing it yourself

The wiki is yours as much as the agents’. The AI Wiki page gives you a file tree and a markdown editor: create, rename, move, and delete pages, with Vim mode if you want it. Seeding pages about your projects by hand is a good way to bootstrap agents’ understanding before their first task.

Last updated on
ProvidersChat