AI
Point an agent at the repository
The authoring loop was built for this. An agent writes the model and the documentation beside it, both as plain files, in one pass — uniform statements it can emit by the thousand, and Markdown it writes like any other prose. There is no connector to build and nothing to import.
Then run it →# one: straight to the repository
$ agent "add the payment gateway as a c4:container"
→ writes models/payments.arch
→ writes descriptions/payments.md
→ reload, and both are in the model
# two: through the Kernel API
$ curl -X POST localhost:7777/api/edits \
-d '{"edits":[{"op":"create", ... }]}'
- No adapter, no connectorThe model is text in a folder. Anything that can write a file can author your architecture.
- The documentation is Markdown, beside itDescriptions live in .md files next to the model. ArchMeta reads them and never rewrites them, so an agent writes the prose in the same pass as the boxes.
- Or drive it through the APIPOST /api/edits takes a batch, guards the round trip and reloads. The same API the Viewer itself uses.
- It is told what it brokeEvery load writes a report of unresolved references and rule violations. That is the feedback an agent needs to fix its own work.