Architecture-as-code for the AI era.
Coding agents can read an entire codebase and reason about its structure — they can describe a system more thoroughly than most humans have the time to. Tecture turns that capability into living architecture documentation.
A useful architecture isn't a single diagram. It's a whole model: every service, datastore, and dependency, at multiple levels of zoom, with each component explained. Existing diagrams-as-code tooling was built for a person to hand-write one diagram in a DSL — the wrong shape for AI authoring, and the wrong shape for understanding a real system.
Tecture defines that model as plain files in your repo: a structured graph as JSON, plus a Markdown description for every component. The Tecture Skill teaches your coding agent the format; from there it authors and maintains the architecture the same way it edits source code. The CLI renders it as an interactive, drill-down explorer in your browser — so anyone on the team can see what was built without reading every file.
1. Install the Tecture skill for Claude Code (one-time)
git clone https://github.com/tecture-io/tecture-skill.git ~/.claude/skills/tecture2. Generate the architecture. In Claude Code, from your project root:
Document this codebase architecture using tecture
3. Render it.
npx @tecture/core
# → http://localhost:3000A small folder of plain files your agent writes and updates with normal fs.writeFile calls. The Tecture Skill packages the format and conventions as reusable instructions for Claude Code; other file-editing agents (Cursor, Copilot, Aider, …) can follow the same instructions directly — no plugin or DSL required.
architecture/
├── manifest.json # top-level diagram + list of diagrams
├── diagrams/
│ ├── system-context.json # nodes + edges for one diagram
│ └── containers.json
└── descriptions/
└── api-server.md # Markdown description for one node
Each diagram is one level of a multi-level architecture view (system → containers → components); each node is a system/service/datastore/etc.; each edge is a relationship like calls, reads, or publishes. See architecture/ for a complete worked example — this repo documents itself.
Contributions welcome. See CONTRIBUTING.md for the dev loop, repo layout, and scripts.
MIT © Tecture.io