Most diagrams are a pain to iterate: changing layout, swapping theme, and producing high-resolution exports typically involve multiple tools and manual steps. Archify flips that flow by letting an LLM-driven agent produce a validated, self-contained HTML/SVG diagram from plain-English prompts, then iterate in chat and export pixel-perfect images without a build chain.
What Sets It Apart
- Agent-first renderer: packaged as a reusable "skill" for Claude, Codex CLI and opencode so agents can generate a typed JSON IR, validate it, render HTML/SVG, run post-render checks, and iterate — all without installing runtime dependencies.
- Dual-theme, single-file exports: exports include a self-contained SVG carrying both dark and light variable sets (uses @media prefers-color-scheme) so a single file follows reader theme on GitHub/Blogs. Raster exports are generated by rasterizing the SVG at up to 4× viewBox to avoid upsampling blur.
- Quality & automation loop: schema validation, layout checks, and an artifact checker prevent common SVG/layout errors; a CLI wraps validate/render/check commands for CI-friendly workflows.
- Practical UX features: theme toggle with keyboard shortcut, copy-PNG-to-clipboard, a small-file self-contained HTML artifact for easy sharing, and real-repo examples that demonstrate the renderer pipeline.
Who It's For & Tradeoffs
Great fit if you want fast, reproducible technical diagrams driven by prompts or integrated into agent workflows — e.g., SREs, architects, developer docs, runbooks, and ML/analytics teams who need quick diagrams tied to narrative text. It shines when you need repeatable, high-quality exports (retina/print) and programmatic checks in CI.
Look elsewhere if you require a full WYSIWYG canvas editor, a large icon runtime of proprietary cloud vendor glyphs, or collaborative multi-user live editing out of the box. Also note some export behaviors depend on browser clipboard/canvas capabilities (WebP and clipboard image writes), and pixel-perfect font rendering for raster exports may need local font installation. The project is MIT-licensed and—at the time of capture—has broad community adoption (several thousand stars), but it intentionally keeps a small footprint rather than bundling a giant icon library.
