Archify

Archify is an open-source agent skill for Cursor, Claude Code, Codex CLI, OpenCode, and Raven that converts natural language system descriptions or repository analysis into interactive, shareable technical diagrams. It is licensed under MIT and currently at stable version v2.13.0, released on August 3, 2026. The project is based on the original Cocoon-AI/architecture-diagram-generator and has accumulated over 9,800 stars on GitHub.

What It Is

Archify is a diagram-as-code agent skill — not a general-purpose drawing editor or a Mermaid theme. You describe your system in chat, and your AI coding agent uses Archify to generate a fully self-contained HTML file containing an interactive, explorable diagram. The output requires zero dependencies, no server, and no runtime: open it in any browser and it works. Diagrams can be refined conversationally ("add Redis", "move auth to the left") without learning a diagram editor.

Five Diagram Types

Archify supports five distinct visual languages, each with its own typed JSON intermediate representation (IR) and validation rules:

• Architecture — system components, cloud resources, databases, caches, services, security groups, and connections; supports AWS/GCP/Azure infra, microservices topology, security boundaries, and network layout
• Workflow — swim-lane processes with semantic nodes, anchored edges, approval gates, async branches, and observability paths
• Sequence — API call chains, request lifecycles, cache fallback paths, auth checks, and async traces
• Data Flow — ETL/ELT pipelines, analytics events, PII isolation, warehouse sync, lineage, and downstream consumers with governance boundaries
• Lifecycle — state machines, object lifecycles, run/order/deployment status transitions with wait states, retries, cancellation, and terminal outcomes

How the Validation Pipeline Works

Archify enforces a generate → validate → deliver pipeline before any artifact is surfaced. Schema, layout, HTML/SVG, route, and label-to-route clearance checks must all pass before a new artifact atomically replaces the last known good output. Failures return machine-readable JSON with stable rule codes, the exact subject, measured evidence, and only supported repair controls — not a Node stack trace or an unstructured retry guess. An optional desktop preview loop watches one JSON source file and keeps the previous verified diagram visible when a save is incomplete or invalid.

Production-Ready Output and Export

Every generated diagram ships with four visual presets (Classic, Signal Flow, Blueprint, Editorial), coordinated dark/light themes, and multiple export formats:

• PNG, JPEG, WebP rasterized natively at up to 4× source resolution — no upsampling blur
• Dual-theme SVG with a @media prefers-color-scheme rule that follows the reader's theme when dropped into a GitHub README
• Browser-native WebM motion recording
• 1200×630 share cards for README, release notes, or social posts
• One-click clipboard copy for instant paste into Slack, Notion, GitHub, or Figma

Agent Integration and Setup

Archify installs in one command via the npx skills toolchain and integrates with Cursor, Claude Code, Codex CLI, and OpenCode through an agent-aware quick-start switcher. Raven supports a manual ZIP install. Once installed, users ask their agent to use Archify by name; the agent generates typed JSON IR, runs validation, and delivers the HTML artifact. Iteration happens in chat with focused natural-language requests, while unrelated diagram structure stays stable between edits.

Update: v2.13.0

The current stable release is v2.13.0, published August 3, 2026. According to the changelog and README, this version includes typed IR across all five diagram modes, real-repository proof, deterministic exact-ID Architecture Delta review (Before/Delta/After snapshots with machine receipts), verified live preview, authored reachability tracing, truthful configurable legends, optional finite motion, guided views, semantic search and relationship exploration, shareable deep links, browser-native WebM recording, explicit standard/showcase quality profiles, and an opt-in deployment ownership contract. The project explicitly notes that automatic Mermaid parsing, general-purpose auto-layout, hosted sharing, and WYSIWYG editing are intentionally outside the current scope.