The AI diagram generator that lives in your repo

Most AI diagram tools are a chat box that hands you an image. Chiltepin turns the prompt into source code: your AI agent writes diagrams as typed YAML blocks inside your Markdown docs, chiltepin check proves they're structurally right, and the build renders them as clean, self-contained SVG. Diagrams get reviewed in pull requests and versioned in git — next to the system they describe.

Text to diagram, verified

One prompt to the agent you already use:

Document our checkout flow in docs/checkout.md — a sequence
diagram of the payment path, an ERD of the order tables, and a
C4 container view. Run chiltepin check and fix anything it reports.

The agent writes the blocks because it has learned the grammar — npx skills add jdiejim/chiltepin -g teaches it once, for Claude Code, Cursor, Codex, Copilot and 70+ other agents. Every block has a strict schema, so the agent's output isn't taken on faith: chiltepin check validates every field and cross-reference, and fails CI when a diagram is wrong. You review content, not syntax.

25+ diagram types, one grammar

Every type below is a typed block with the same YAML shape — no per-diagram DSL to learn, for you or the model:

• Sequence
• ERD
• Flowchart
• State machine
• C4
• UML
• Data flow
• Swimlane
• Architecture map
• Dependency graph
• Gantt
• Timeline
• Git graph
• Sankey
• Treemap
• User journey
• Wireframe
• Quadrant

See every one rendered live in the block catalog, or the guide to picking the right diagram type — a growing library of typed blocks, because documents are more than diagrams: API endpoints, decision records, roadmaps, and slide decks come from the same pipeline.

Why not just generate an image?

• Deterministic. The same block always renders the same SVG — at build time, with zero runtime JavaScript.
• Editable. Tweak a label in Studio or any text editor; re-prompt the agent for bigger changes. No re-rolling an image and hoping.
• Diffable. An architecture change shows up as a three-line diff in review, not a new PNG.
• Checkable. A schema means CI can catch the diagram that references an actor that doesn't exist — before it ships.

AI architecture diagrams from your actual codebase

Because the agent generating the diagrams is the same one that can read your code, the diagrams describe the system as it is — the request flow from the real handlers, the ERD from the real migrations. That's the core of the docs-as-code workflow: documentation lives with the code, changes with the code, and is checked like the code. If you're coming from a diagram DSL, see how it compares to Mermaid.