Archify is an agent skill that turns a written description, or an existing repository, into an interactive architecture diagram shipped as a single self-contained HTML file. It is meant for developers who have to explain a system to somebody else — a reviewer on a pull request, a new teammate on their first day, a design document, a talk — and who do not want to hand-drag boxes in a canvas tool every time the code moves. The project is MIT-licensed, written mostly in JavaScript, and the video notes it gained roughly 48,000 stars in a single month.
What it does
The premise in the README is blunt: most diagrams go stale the moment the code changes. Archify attacks that by making the diagram cheap to regenerate. You describe what you want to understand, plan, or share, your coding agent runs the skill, and the result is one HTML file you can open, explore, and send to anybody.
The diagram types it covers, per the project's own description, are architecture, workflow, sequence, data-flow, and lifecycle. The output is not a flat image:
- nodes are clickable — pick one and the whole downstream path is traced for you
- themes can be switched, and the view zooms
- there is motion in the rendering rather than a static snapshot
- export is crisp, so the same file feeds a slide or a document
The README is deliberately broader than software architecture. It pitches the same skill for travel itineraries and learning maps — anything you want to lay out visually — and points at a public gallery of live demos plus a scenario guide so you can see what other people built before writing a prompt.
How it works
Archify is a skill, not an application you install and click around in. Your agent reads its instructions and writes the HTML; the skill's value is the rules and the templates it imposes, so you get a consistent, readable diagram instead of whatever the model would have improvised.
The part worth paying attention to is verification. When the source is a real codebase, the skill checks every box in the diagram against the actual files before presenting it, rather than rendering a plausible-looking picture of a system that does not exist. That is the difference between a diagram and a guess, and it is the main claim the project makes for itself.
It is not tied to one vendor. The repository's topics list Claude Code, Codex, opencode, and a DeepSeek harness plugin, so the same skill is meant to travel across the coding agents people already use.
Getting started
You need a coding agent; Archify plugs into one rather than replacing it. The fastest way to judge whether it fits your work is to open the hosted gallery of live demos linked from the README and look at the finished HTML — everything the skill can do is visible there in a browser, no setup required. The README then has its own get-started section for wiring the skill into your agent, plus a scenario guide that maps common situations to prompts. Translated READMEs are provided for Chinese and Japanese readers.
When to use it / when not
It fits well when a diagram is a means to an end and its value decays fast: onboarding notes, a pull-request explanation, a design review, or a first orientation pass over a repository you inherited and do not yet understand. Regenerating beats maintaining in all of those cases.
It fits badly in a few situations. If you have no agent set up, or no budget for the tokens a repository walk costs, there is nothing here for you. If the diagram must live inline in a README on GitHub, a self-contained HTML page is the wrong artifact. And if you need a small, reviewable, deterministic source file that a build pipeline regenerates the same way every time, a fixed diagram syntax is a better match than a prompt.
Alternatives
The obvious comparison, and one the project makes itself by claiming the mermaid-alternative label, is the established diagram-as-code family: Mermaid, PlantUML, Graphviz. Those take a terse text source, render deterministically, and are supported almost everywhere, including directly in GitHub for Mermaid — but you write the source yourself, and the output is a static picture. On the other side sit manual canvases such as Excalidraw or draw.io, which give you total control and no automation at all. Archify sits between them: the input is prose, the work is done by an agent, and the output is interactive.
Take this repository seriously if you maintain a system that other people keep asking you to explain, and you already run a coding agent daily. The verification step is what earns it a place in your workflow — a diagram you can trust against the current files is worth far more than a prettier one you cannot. If you want a stable, checked-in diagram source that never depends on a model, stay with the text-based tools.