GitHub Trending This Week: Archify — Architecture Diagrams as Code for AI Agents
Archify (tt-a1i/archify) is the top repo on GitHub’s trending list today — a JavaScript project sitting at 28,430 stars (as of August 29), MIT-licensed, and actively developed (latest tag v2.15.0, August 17 2026; the dev line is already at v2.16.0-dev.0). It is still #1 on GitHub’s trending list, and the star count keeps climbing — up about 80 stars in the last half hour. In one sentence: it is an agent skill that turns a codebase or system description into a polished, interactive system map directly in chat.
If you are tired of architecture diagrams that no longer match the code — or of asking an AI for a “simple flowchart” and getting something that looks plausible but is entirely made up — Archify is worth a closer look. The interesting part is not the pretty SVG. It is the validation layer underneath.
1. What Archify does
Archify is a Node.js rendering and validation system made for coding agents: Cursor, Claude Code, Codex CLI, and OpenCode. The agent produces a typed JSON intermediate representation of the system you describe; Archify then deterministically compiles that JSON into HTML and SVG.
The output is meant to survive outside the chat: one self-contained HTML file, plus exports as PNG, SVG, WebM, and a canonical 1200×630 share card (useful for READMEs, releases, or PR descriptions). A single click switches between dark and light themes.
2. Five diagram types, chosen for the job
Archify is not a one-size-fits-all “draw me a diagram” tool. Each of the five types has a schema and a purpose:
| Type | Best for |
|---|---|
| Architecture | Components, services, storage, trust boundaries |
| Workflow | CI/CD, approvals, tool calls, runbooks |
| Sequence | API calls, cache fallback, auth, async traces |
| Data Flow | Pipelines, lineage, PII, consumers |
| Lifecycle | States, retries, waits, terminal outcomes |
When you ask for one, you give it scope — e.g. “Show 8–12 core components, one primary path, external dependencies, and trust boundaries.” The agent chooses hierarchy, spacing, routes, and emphasis; the rendering spreads connection endpoints deterministically instead of piling arrows onto one midpoint.
3. The part that matters: validation before delivery
This is the feature that separates Archify from a Mermaid clone. Before any artifact replaces the last known good output, atomic validation must pass: schema checks, layout rules, HTML/SVG checks, route checks, and label-to-route clearance. Failures come with a machine-readable repair receipt — validate --json and deliver --json return stable rule codes, the exact subject, measured evidence, and only supported repair controls. No Node stack traces, no unstructured retry guesses.
The optional preview mode watches one JSON file and refreshes only verified revisions; if a save is incomplete, the previous verified diagram stays visible. It binds only to 127.0.0.1 on a random port and adds no runtime to the generated HTML.
Interaction is designed to stay truthful: focus, upstream/downstream reach, route probing, role comparison, and guided “stories” reuse authored nodes and relationships — they never invent topology or claim runtime impact. On top of that, an Evidence-backed mode marks nodes with SRC n and opens Git-verified file ranges pinned to a public commit.
For production deployment reviews, the optional deployment-ownership profile goes further: it fails closed when owners, single-region placement, private database scope, or named boundary crossings are missing — and it is never enabled silently.
4. Comparing architecture changes
There is even a built-in review flow. node archify/bin/archify.mjs compare architecture base.json head.json architecture-delta.html --json produces a Before / Delta / After comparison of two validated snapshots, with exact added, removed, changed, moved, or rerouted facts and a machine receipt — the kind of thing that makes a PR review section actually readable.
5. How to get started
Installing takes one command:
npx skills add tt-a1i/archify -g
Then ask your agent:
Use archify to map this repository’s runtime architecture.
Want to try it without installing? npx skills use tt-a1i/archify@archify --agent codex runs it on demand. The bundled CLI also has zero dependencies and works standalone: node bin/archify.mjs guide "Show CI/CD checks, approval, deploy, and rollback" or validate/deliver against a JSON file for full control.
Beyond the four main agents, there is a community DeepSeek Harness plugin (dsh plugin --profile web add @tt-a1i/archify-dsh@0.1.0) and a manual ZIP install for Raven. The project page and a “Proof Lab” gallery with eleven checked-in scenarios and their validation receipts are linked from the README.
6. Who it is for
Most immediately: engineers who already run coding agents and need to understand a codebase they did not write — new teammates, inherited projects, or a repo you touched last six months ago. The ability to trace a route (e.g. the exact Web App → Postgres path) or compare roles (backend vs. database) makes the output useful in a code review, not just on a slide.
If you are newer to running agents locally, our guides on getting started with Hermes Agent and installing and using OpenClaw cover the broader agent landscape — Archify is the kind of skill you would add to one of those setups.
7. Bottom line
Archify is a narrow, well-scoped tool with an unusually strong commitment to honesty: diagrams are compiled from typed, validated sources, failures are reported in machine-readable receipts, and interactions never invent topology. If your frustration with AI-generated diagrams is that they look great but can’t be trusted, this is exactly the project to try.
A taste of a first run against an existing repo:
Use archify to draw this login flow: Browser -> Web App -> API -> JWT validation ->
Redis session lookup -> PostgreSQL fallback. Keep the cache-miss path secondary.
That is the whole workflow: prompt, validate, deliver, iterate.
Hero image: Ted Goldman, Public domain, via Wikimedia Commons.
Frequently Asked Questions
Is Archify free?
Yes. Archify is released under the MIT License, which allows free use, modification, and distribution. The skill installs locally through npm — there is no hosted service or paywall.
Which AI coding agents does Archify work with?
Cursor, Claude Code, Codex CLI, and OpenCode are the officially supported surfaces, installed via `npx skills add tt-a1i/archify -g`. There are also community integrations for DeepSeek Harness (a plugin) and Raven (a manual zip install into its skills folder).
Does Archify replace Mermaid?
No — and the project says so explicitly: it is not a Mermaid theme or a general-purpose drawing editor. Archify renders typed, validated system descriptions into interactive HTML/SVG diagrams; automatic Mermaid parsing and WYSIWYG editing are intentionally out of scope.
What diagram types can it produce?
Five: Architecture (components, services, storage, trust boundaries), Workflow (CI/CD, approvals, runbooks), Sequence (API calls, cache fallback, auth, async traces), Data Flow (pipelines, lineage, PII, consumers), and Lifecycle (states, retries, waits, terminal outcomes).