Prompt-to-diagram
Describe a diagram in plain English and a frontier model — Claude, GPT, Gemini, or any OpenAI-compatible endpoint — turns it into a laid-out, fully editable canvas.
Most diagramming tools produce dead artifacts — stale the moment the system changes. Sutra draws semantic, living diagrams: typed components you can bind to a source of truth, laid out for you by AI from a plain-English description.
Describe a diagram in plain English and a frontier model — Claude, GPT, Gemini, or any OpenAI-compatible endpoint — turns it into a laid-out, fully editable canvas.
Iterate conversationally — "add a payments service", "make the error path explicit". Each edit applies to the live diagram and can be reverted with one click.
Keys stay in your browser and prompts go straight to the provider you chose — there's no Sutra server in between. No account, no tracking, nothing stored remotely.
Architecture graphs, flowcharts, swimlanes, ER diagrams, mind maps, user journeys, and Gantt timelines — for engineering and product audiences alike.
"Explain this architecture", single points of failure, fan-in hotspots, orphans, unbound components — computed locally over the semantic graph, no AI required.
Full-screen presentation mode on F5, and downloads with control over padding, scale, and background: PNG, SVG, or a re-importable .sutra.json you can share.
"Swimlane for client, middleware and server — client requests product info." One sentence in, a laid-out diagram out. Or skip AI entirely and build from scratch.
Drop typed components from the palette, drag to connect, tidy up with auto-layout, undo with Ctrl/Cmd+Z, and reach everything from the Ctrl/Cmd+K command palette.
Present full screen, or export PNG, SVG, or .sutra.json — the recipient can keep refining with their own key, or none at all.