JT Draw

JT Draw is an AI-driven SVG asset generator. Users describe what they need in plain language — charts, diagrams, icons, illustrations — and get a polished, editable SVG in seconds. No design skills required, no complex tools to learn.

The studio interface is simple: a prompt bar at the bottom, a live canvas above. Type what you want, the AI generates it, refine with follow-up prompts until it's right.

Creating professional vector assets typically requires either design expertise (Figma, Illustrator) or developer time (hand-coding SVG, wiring up D3.js). Neither is fast, and both create bottlenecks. Teams need charts for presentations, icons for UIs, diagrams for docs — and they need them now, not next sprint.

The core technical challenge: making AI-generated SVGs reliable enough for production use. Raw LLM output is inconsistent — missing viewBoxes, broken markup, unsafe content. A single bad generation erodes trust in the tool.

A self-correcting agent engine sits between the user and the AI. Claude generates raw SVG markup from natural language prompts. The server extracts, validates, and sanitizes the output — checking for well-formed markup, proper viewBox attributes, and stripping any unsafe content (scripts, event handlers). If validation fails, the agent automatically retries with error context, up to a category-aware budget (2 rounds for icons, 4 for complex illustrations).

The entire flow is server-rendered. SVGs are generated via the Claude API, validated server-side, and delivered to the canvas via HTMX swap. No client-side JS framework — just the SVG and a prompt bar.

• Natural language to SVG — describe charts, diagrams, icons, or freeform illustrations in plain English
• Self-correcting agent loop — automatic retries with error context when generation fails validation
• Production-safe output — sanitization strips scripts, event handlers, and unsafe URIs from every generated SVG
• Category-aware generation — iteration budgets tuned per asset type (icons are fast, illustrations get more attempts)
• Instant preview — server-rendered SVG appears in the canvas immediately, no loading spinners
• Freemium model — free tier with generation limits, Pro tier for power users

• Zero client-side framework — SSR with Fastify + fluent-html, interactivity via HTMX morphing
• Direct SVG generation — AI outputs raw SVG markup, no intermediary DSL or code execution layer
• Validation pipeline — extraction, sanitization, and structural checks before any SVG reaches the client
• Multi-turn conversation — agent maintains context across retries for coherent error correction
• Google SSO — single sign-on, no password management overhead