glacius-labs/markitect

Markitect keeps one canonical model of your project in Markdown, checks it like code, and has a team of AI agents derived from its architecture realize and check every change.

Why · How it works · Quick start · Example · Status · Documentation

[!IMPORTANT] Markitect is in active development and runs from source; there is no release yet. Every command on this page exists on main. Whether its Runs keep a project at least as consistent as one strong agent with ordinary documentation, with fewer defects left for people to repair, is the question its evaluation answers; a pilot has run, and the counted series follows (status).

Why Markitect

When AI agents build most of a project, its design, rules, documentation, code, tests and configuration drift apart. A decision about behavior reaches the code, the tests, the documentation and the agents’ own instructions. An agent finishes its task and misses a rule two folders away. Consistency passes take hours and find more each time, and a pull request that one AI wrote, another reviewed and a person approved on trust carries the gap to main.

Markitect moves the knowledge of what belongs together out of the agents’ memory and into one place you own: the canonical model. You describe the project once, as typed documentation in Markdown. Code, tests, configuration and documentation are its realizations. When intent changes, you change the model; Markitect works out what the change reaches, has the responsible agents bring the repository in line, checks every step independently and lands the result with a record of every verdict.

Features

• Typed documentation. One Markdown file per Definition, its Kind in the front matter, links by path such as [[members/member]]. Projects define their own Kinds; the standard Libraries std, rest and cli ship in the binary.