Archify

Generate beautiful architecture, technical workflow, sequence, data-flow, and lifecycle diagrams in chat. Switch dark / light. Copy to clipboard or export crisp up-to-4× PNG / JPEG / WebP / SVG.

Archify is an agent skill for Claude, Codex CLI, and opencode. It turns a plain-English description of your system or process into a polished, self-contained technical diagram — a single HTML file you can open, toggle themes on, copy to the clipboard, and export at maximum resolution.

• No design skills needed — describe your architecture in English, get a diagram
• Workflow, sequence, data-flow, and lifecycle diagrams too — technical flows, approvals, tool calls, CI/CD, runbooks, request call chains, data pipelines, PII boundaries, and state machines can be drawn
• Built-in theme toggle — one click between dark and light, persists across sessions
• Copy PNG to clipboard — one click, paste straight into Slack / Notion / GitHub
• Ultra-crisp image export — PNG / JPEG / WebP rendered natively at up to 4× source resolution (no upsampling blur), or SVG for true vector
• SVG follows system dark/light — exported SVGs ship with both variable sets + @media (prefers-color-scheme), so dropping one into a GitHub README makes it follow the reader's color preference (no more two PNGs wrapped in <picture>)
• Self-contained HTML — the generated file has zero dependencies, share by sending it
• Iterate by chat — "add Redis", "move auth to the left", "use emerald for the API"

Project page: tt-a1i.github.io/archify

中文

Preview

Same diagram, two themes, one click to switch:

Dark	Light

The Export menu — Copy PNG to clipboard plus 4 download formats (all raster exports at up to 4× source resolution):

Live example: examples/web-app.html — open in a browser, press T to toggle theme, E to open Export. Append ?theme=light or ?openExport=1 to the URL for deterministic screenshots.

Diagram types

Archify now has five primary outputs:

Type	Good for	How to ask
Architecture	System components, cloud resources, databases, caches, services, boundaries, security groups	Describe the system structure
Workflow	Request lifecycles, approval flows, tool calls, CI/CD, runbooks, incident response	Describe participants, step order, and key branches
Sequence	API call chains, request lifecycles, cache fallback, auth checks, async trace, service interactions	Describe who calls whom, in what order, and what returns
Data Flow	Data pipelines, ETL/ELT, analytics events, PII isolation, warehouse sync, lineage, downstream consumers	Describe sources, processing stages, storage, sensitivity boundaries, and consumers
Lifecycle	State machines, order/task/deployment/agent-run lifecycles, wait states, retries, cancellation, timeout, terminal states	Describe states, transition events, retry paths, and terminal outcomes

Workflow is not trying to replace every general-purpose flowchart. It is a technical communication diagram: swimlanes, semantic colors, a clear happy path, and secondary async / approval / trace paths.

Use archify to draw a workflow:
User submits a request -> Agent plans -> Approval Gate if needed -> Tool Call -> Trace Log -> Final Reply

Open the example here: examples/workflow-agent-tool-call-rendered.html.

Sequence diagrams explain a narrower interaction over time:

Use archify to draw a sequence diagram:
User opens a page, the frontend calls the API, the API verifies JWT, reads Redis, falls back to Postgres on cache miss, returns JSON, and emits trace.

Open the example here: examples/sequence-cache-miss-request.html.

Data Flow diagrams explain how data assets move and change:

Use archify to draw a data flow:
Web and Mobile emit analytics events, Edge API collects them, Consent Gate filters PII, Kafka carries accepted events,
Warehouse stores analytics tables, Feature Store derives daily features, Dashboards and an ML Model consume downstream data.

Open the example here: examples/dataflow-product-analytics.html.

Lifecycle diagrams explain how an object changes state:

Use archify to draw a lifecycle diagram:
Agent Run starts at Queued, moves through Planning, Executing, and Reviewing. It can pause at Needs Approval,
wait at Blocked, retry after Failed, end at Cancelled or Expired, or finish at Completed.

Open the example here: examples/lifecycle-agent-run.html.

What's new

Archify is based on Cocoon-AI/architecture-diagram-generator v1.0 (dark-only, HTML output). 2.0 rewrote the template around a themeable CSS-variable system and added a client-side export pipeline. 2.1 added copy-to-clipboard + keyboard nav. 2.2 added a print stylesheet + local-font fallback. 2.3 fixed a long-standing upsampling bug and made every raster export genuinely sharp at 4× source resolution (the 1× / 2× / 4× selector introduced in 2.1 was removed at the same time — it only encouraged picking a soft-looking scale). 2.4 upgraded the SVG export to a dual-theme self-contained file — drop the same .svg into a GitHub README and it follows the reader's dark/light preference automatically. 2.5 is a hardening release: a batch of verified renderer and validation bug fixes (template slot corruption, cross-lane lifecycle overlap detection, light-theme lane colors in PNG/SVG exports, Safari clipboard copy), LLM-ergonomic validation errors with numeric thresholds and fix suggestions, Mermaid accepted as an input dialect (mapped through SKILL.md prompts, not a parser), CJK-aware text measurement plus CJK font fallbacks, and a golden-file test suite with CI.

Feature	v1.0	2.0	2.1	2.2	2.3	2.4	2.5
Dark theme	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Light theme	—	Toggle	Toggle	Toggle	Toggle + T shortcut	Same	Same
PNG / JPEG / WebP download	manual screenshot	2× bitmap-upsampled	1× / 2× / 4× selector (still upsampled)	same	4× rasterized natively — no blur	Same	Light-export lane colors fixed
SVG download	—	Vector, styles inlined (single theme)	Same	Same	Same	Dual-theme self-contained (@media prefers-color-scheme)	Same (lane colors fixed)
Copy PNG to clipboard	—	—	Yes	Same	Yes (up to 4×)	Same	Safari fix
Keyboard shortcuts	—	—	T / E + menu nav	Same	Same	Same	Same
Accessibility	—	—	ARIA + focus-visible	Same	Same	Same	Same (+ menu a11y fixes)
Print stylesheet	—	—	—	Yes	Yes (+ landscape + 2-col cards)	Same	Same
Local-font fallback on export	—	—	—	Yes	Yes	Same	+ CJK font fallback
Styling model	Inline fill / stroke	CSS custom properties + semantic classes	Same	Same	Same	Same	Same
Typed renderers + schema validation	—	—	—	—	—	—	Yes (with tests & CI)

Quick start

1. Install the skill

Archify is packaged as an agent skill directory (archify/SKILL.md), so the same archify.zip works with Claude, Codex CLI, and opencode.

Claude.ai:

1. Download archify.zip
2. Go to Settings -> Capabilities -> Skills
3. Click + Add and upload the zip file
4. Toggle the skill on

Claude Code CLI:

# Global (all projects)
unzip archify.zip -d ~/.claude/skills/

# Or project-local
unzip archify.zip -d ./.claude/skills/

Codex CLI:

# Global (all projects)
unzip archify.zip -d ~/.agents/skills/

# Or project-local
unzip archify.zip -d ./.agents/skills/

opencode:

# Global (opencode-native)
unzip archify.zip -d ~/.config/opencode/skills/

# Or project-local
unzip archify.zip -d ./.opencode/skills/

# Also works: the shared agent directory used above for Codex
unzip archify.zip -d ~/.agents/skills/

The typed renderers (workflow / sequence / dataflow / lifecycle) rely on ajv for schema validation, which takes a one-time npm install in the installed skill directory. Many agent runtimes can run it automatically on first use, following the Setup instructions in SKILL.md — or run it yourself:

cd ~/.agents/skills/archify && npm install

Without the dependency the renderers skip schema validation (layout checks still run).

Claude.ai Projects (alternative): Upload archify.zip to your Project Knowledge.

What each install surface can do:

Install surface	Capability
Claude Code	Full — runs the typed renderers + schema validation
Codex CLI	Full — install to ~/.agents/skills/ or .agents/skills/
opencode	Full — install to .opencode/skills/, .agents/skills/, or another supported skills directory
Claude.ai (zip upload)	Usually full — depends on whether the sandbox can npm install, which it typically can
Project Knowledge	Architecture mode only — no code execution, purely prompt-driven

2. Describe your system

Any of these work:

Have AI analyze your codebase:

Analyze this codebase and describe the architecture. Include all major
components, how they connect, what technologies they use, and any cloud
services or integrations. Format as a list for an architecture diagram.

Write it yourself:

- React frontend talking to a Node.js API
- PostgreSQL database
- Redis for caching
- Hosted on AWS with CloudFront CDN

Or ask for a typical architecture:

What's a typical architecture for a SaaS application?

3. Ask Claude to use the skill

Use your archify skill to create an architecture diagram from this description:

[PASTE YOUR ARCHITECTURE DESCRIPTION HERE]

That's it. Claude generates an HTML file you can open in any browser. Iterate by chat: "add Redis", "swap Postgres for MySQL", "highlight the auth path".

Using the output

Open the generated HTML in any browser. Top-right you'll see two buttons:

• Theme button (Dark / Light) — one click flip, persisted across sessions. Shortcut: T.
• Export menu — opens a dropdown with five actions (Copy PNG + download PNG / JPEG / WebP / SVG). Shortcut: E.

Export menu

Action	What it does
Copy PNG	Puts a PNG of the current diagram straight on your clipboard. Paste into Slack, Notion, GitHub, Figma.
Download PNG / JPEG / WebP	Saves a raster image. JPEG/WebP are painted over the current theme's background (no alpha); PNG keeps transparency.
Download SVG	Vector export with all styles inlined, dual-theme self-contained. The file ships with both dark and light CSS variable sets plus a @media (prefers-color-scheme) rule — drop the same .svg into a GitHub README or blog and it follows the reader's preference automatically. Still editable in Figma / Illustrator. Scales losslessly.

Every raster export (Copy PNG, Download PNG/JPEG/WebP) is rendered natively by the browser at up to 4× the diagram's intrinsic resolution (oversized diagrams step down to 3×/2× to stay within browser canvas limits) — the serialized SVG is given a width/height of 4 × viewBox, rasterized by the browser at that resolution, and drawn to the canvas at natural size (no upsampling). This produces genuinely crisp output for retina displays, slides, and print. There is no scale dial — maximum safe sharpness is always chosen automatically.

Keyboard

• T anywhere — toggle theme
• E anywhere — open the Export menu
• ↑ ↓ inside the menu — navigate actions
• Home / End — jump to first / last action
• Enter / Space — activate
• Esc — close menu

URL parameters

• ?theme=light or ?theme=dark — force a starting theme (deterministic screenshots, share links, embeds)
• ?openExport=1 — auto-open the Export menu on load (demo / docs screenshots)

Notes

• WebP support depends on your browser's canvas encoder. If unavailable (older Safari), the menu item is dimmed and disabled. PNG and JPEG are universal.
• Clipboard support for images requires ClipboardItem + navigator.clipboard.write (Chromium, Firefox 127+, Safari 16+). If unavailable, Copy PNG is dimmed.
• Fonts in exports: raster images use the system monospace fallback (ui-monospace / Menlo / Consolas) because the sandboxed image-rendering context can't fetch Google Fonts. Install JetBrains Mono locally for pixel-perfect rendering.

Example prompts

Web app:

Create an architecture diagram for a web application with:
- React frontend
- Node.js/Express API
- PostgreSQL database
- Redis cache
- JWT authentication

AWS serverless:

Create an architecture diagram showing:
- CloudFront CDN
- API Gateway
- Lambda functions (Node.js)
- DynamoDB
- S3 for static assets
- Cognito for auth

Microservices:

Create an architecture diagram for a microservices system with:
- React web app and mobile clients
- Kong API Gateway
- User Service (Go), Order Service (Java), Product Service (Python)
- PostgreSQL, MongoDB, and Elasticsearch databases
- Kafka for event streaming
- Kubernetes orchestration

Data flow / product analytics:

Use archify to draw a data flow:
- Web App and Mobile SDK produce clickstream events
- Edge API collects events
- Consent Gate filters identity and PII
- Kafka/Event Stream carries accepted events
- Warehouse stores normalized facts
- Feature Store derives daily feature vectors
- Dashboards and ML Model consume downstream data

State machine / lifecycle:

Use archify to draw a lifecycle diagram:
- A task starts at Queued
- Planning builds the plan
- Executing calls tools
- Reviewing checks quality
- Needs Approval and Blocked are wait states
- Failed can retry, while Cancelled / Expired / Completed are terminal states

Color palette

Component Type	Color	Use for
Frontend	Cyan	Client apps, UI, edge devices
Backend	Emerald	Servers, APIs, services
Database	Violet	Databases, storage, AI/ML
Cloud / AWS	Amber	Cloud services, infrastructure
Security	Rose	Auth, security groups, encryption
Message Bus	Orange	Kafka, RabbitMQ, SNS, event buses
External	Slate	Generic, external systems

Each color has coordinated dark-mode and light-mode variants that switch together via the theme toggle.

Technical details

• Styling: CSS custom properties on :root + [data-theme="light"]; SVG elements reference semantic classes (c-frontend, t-muted, a-emphasis, etc.). Toggling data-theme on <html> re-themes the entire diagram including gradient, grid, arrows, and mask rects.
• Export pipeline: The SVG is cloned, host <style> is inlined, and current theme variables are resolved and written into a :root rule on the clone. For raster formats the clone's width/height are set to 4 × viewBox so the browser rasterizes the vectors at target resolution natively; the canvas is sized to match and the image is drawn at natural size (no bitmap upsampling). toBlob(mime) then produces the file. JPEG gets an explicit background fill since it has no alpha. If the target resolution would exceed the browser's canvas limits, the pipeline automatically falls back to 3× or 2× so the export still succeeds.
• Self-contained output: Single HTML file, Google Fonts link + inline SVG + ~19 KB of embedded JS. No build step, no JS framework, no server — the generated HTML itself has zero dependencies (the typed renderers need npm install for ajv, see Install).
• Browser support: Any modern browser (Chrome, Safari, Firefox, Edge). Needs Image + canvas.toBlob with image/webp support for WebP export.

Attribution

Archify is a fork / rewrite of Cocoon-AI/architecture-diagram-generator (MIT, v1.0) by Cocoon AI. The original's clean visual design — color palette, grid background, summary-card layout, JetBrains Mono typography — is preserved. All credit for the original aesthetic belongs to them.

Archify 2.x contributes:

• Refactor of the template to a CSS-variable theme system (dark + light)
• Theme toggle + localStorage persistence + prefers-color-scheme default
• Built-in PNG / JPEG / WebP / SVG export menu + copy to clipboard
• 4× native rasterization (fixes upsampling blur)
• Dual-theme self-contained SVG export (single file follows the host's prefers-color-scheme)
• Keyboard navigation + accessibility semantics
• Print stylesheet + local-font fallback
• Updated SKILL.md to guide Claude toward class-based (themeable) diagrams

Both projects are MIT-licensed.

Roadmap

See ROADMAP.md.

Next up is v3.0 — JSON IR stabilization: a minimal diagram.json intermediate format so Claude can make local coordinate edits without drifting unrelated components, with git diff-friendly output and theme/palette swaps that don't require re-rendering.

About Mermaid import: the automatic Mermaid parser route was cut after an experiment showed that auto-layout + archify CSS doesn't look meaningfully better than native Mermaid (experiments/v3-mermaid-validation/RESULT.md). Archify's aesthetic core is Claude's layout judgment, not the CSS. You can still paste Mermaid code and have Claude lay out an archify-style diagram from scratch — it goes through the SKILL.md prompts, not a parser.

The former v2.4 / v2.5 plans (?exportScale=N, color-blind palettes, share links) were also dropped. See the ROADMAP "Not planned" section for the rationale.

License

MIT — free to use, modify, and distribute.

Contributing

Issues, PRs, and shared diagrams welcome.