Archify Setup: AI Diagrams for Claude Code in Minutes

Archify is a free, MIT-licensed skill that turns a plain-English description or an entire codebase into an interactive, verifiable architecture diagram, and it just landed at number 5 on GitHub's daily trending page with over 1,000 stars added in a single day on top of 17,600-plus total. Instead of opening a whiteboard tool, you type a request in Claude Code, Cursor, Codex CLI, or opencode, and the agent hands your request to Archify, which generates a typed diagram source, validates it, and writes out a self-contained HTML file you can open, present, and share.

• Install is one command, npx skills add tt-a1i/archify -g, and it works the same way on Windows, macOS, and Linux because it only needs Node.js 18 or newer.
• It supports five diagram types: architecture, workflow, sequence, data flow, and lifecycle, all generated from typed JSON rather than freehand SVG.
• Every diagram passes schema, layout, and route validation before it is written to disk, and a failed candidate never overwrites your last good diagram.
• Output is one portable HTML file with dark and light themes, keyboard-driven exploration, and one-click exports to PNG, SVG, WebM, and a 1200×630 share card.

What is Archify and why is it trending?

Archify describes itself as an agent skill, not a standalone app: it plugs into an AI coding agent you already use and gives that agent a disciplined way to produce diagrams instead of inventing SVG from scratch. Under the hood it works from a typed JSON intermediate representation, so every diagram has a reproducible source you can hand-edit or regenerate. The project is maintained by tt-a1i, ships MIT-licensed, and its latest tagged release is v2.15.0. The trending spike tracks a real gap: developers increasingly lean on AI agents to explain unfamiliar codebases, and a wall of text explaining a system is worse than a picture of it. Archify's pitch is that the picture should be checked before you trust it, not just generated and hoped for.

RelatedSuperpowers Setup: Give Your AI Coding Agent a Real Workflow

How do you install Archify on Windows, macOS, and Linux?

Archify installs through the community skills CLI, which runs on Node.js and behaves identically across operating systems. Open a terminal (PowerShell on Windows, Terminal on macOS or Linux) and run:

# install Archify globally as an agent skill
npx skills add tt-a1i/archify -g

That single command works for Claude Code and opencode out of the box. Cursor needs an explicit, non-interactive flag set if you are scripting the install rather than clicking through prompts:

# explicit, non-interactive install for Cursor
npx -y skills add tt-a1i/archify --skill archify --agent cursor --global --copy --yes

To try it once without a permanent install:

# one-off trial run, no global install
npx skills use tt-a1i/archify@archify --agent codex

Under the hood, the skill lands in a plain folder your agent already scans: ~/.claude/skills/ or .claude/skills/ for Claude Code, ~/.agents/skills/ or .agents/skills/ for Codex CLI, and ~/.config/opencode/skills/, .opencode/skills/, or .agents/skills/ for opencode. If your agent isn't on that list, you can still use Archify manually: download archify.zip from the Releases page and upload it under Claude.ai's Settings → Capabilities → Skills, or attach it to a project's Knowledge base for a prompt-driven fallback.

How do you generate your first diagram?

Once the skill is installed, you do not run a separate command to draw anything: you just ask your agent in plain language, and it invokes Archify for you.

# ask your agent, in chat, after installing the skill
Analyze this repository, then use archify to create a high-level
runtime architecture diagram. Show 8-12 core components, one
primary path, external dependencies, and trust boundaries.

Archify also ships a zero-dependency CLI inside the installed folder for testing, previewing, and scripted use, useful if you want to sanity-check the tool before trusting it with a real repo:

RelatedHallmark Setup: Kill AI Slop Design in Claude Code

# from inside the installed archify/ folder
node bin/archify.mjs doctor
node bin/archify.mjs demo /tmp/archify-demo
node bin/archify.mjs guide "Show CI/CD checks, approval, deploy, and rollback"

The guide command is worth running early: it is a small local advisor that recommends which of the five diagram types (architecture, workflow, sequence, data flow, or lifecycle) fits a description you type, before you spend agent turns generating the wrong shape. Once a diagram exists, an optional local preview mode watches the JSON source and live-reloads the rendered HTML, binding only to 127.0.0.1 and keeping the last verified version visible if a new edit fails validation.

How does Archify compare to Mermaid and hand-drawn diagrams?

Mermaid is faster for a five-node sketch you can type yourself, and Archify explicitly does not auto-parse Mermaid syntax or act as a general-purpose drawing editor. Where Archify earns its place is anything with real complexity: a service map with a dozen components, a cache-miss sequence, or a before-and-after comparison of an architecture change, where hand-placing boxes in Figma is slow and Mermaid's auto-layout starts piling arrows on top of each other.

What are the gotchas before you rely on it?

Three things worth knowing up front. First, Archify is not a drawing tool you operate directly: you need a supported agent (Claude Code, Cursor, Codex CLI, or opencode) doing the prompting, so it is not useful to someone without one of those already installed. Second, it deliberately stays out of scope for automatic Mermaid parsing, general-purpose auto-layout, hosted sharing, and WYSIWYG editing, so do not expect it to replace a full design tool. Third, the optional deployment-ownership review profile for architecture diagrams fails closed on purpose: if owners, region placement, or boundary crossings are missing from what you described, it will not silently guess, it flags the gap instead. That is a feature for anyone using diagrams in a production review, but it means an incomplete prompt gets rejected rather than filled in with invented detail.

Our take

The interesting decision in Archify is not the diagram output, it is the validation gate in front of it. Plenty of tools will happily hand an AI agent a blank SVG canvas and let it improvise boxes and arrows; Archify instead forces every diagram through a typed schema and layout checker before anything reaches disk, and keeps the last good version visible when a new attempt fails. That is the right instinct for a tool meant to be trusted in a design review or an onboarding doc, where a subtly wrong architecture diagram is worse than no diagram at all. The one-command install and the plain-English prompting make it cheap to try, and the zero-dependency CLI means you are not locked into needing an agent running at all times to poke at an existing diagram. If you already work inside Claude Code, Cursor, or Codex CLI, this is a low-risk five-minute add that pays off the first time you need to explain a real system to someone else.

Original analysis by GenZTech. Tool documentation: tt-a1i/archify on GitHub.