Demos

What an AI can render

One page per render tool an agent can call. Each loads a real scene through the surface-client pipeline — the same renderer a live session uses.

Sankey

Flow, leakage, allocation — the agent supplies nodes and links, the server composes the whole editorial Sankey with in-place annotations.

Ranking and current-vs-target — ranked bars with optional contrasting target ticks, thesis, takeaway and supporting metric.

Part-of-whole breakdowns — squarified cells labelled with category, value, and share. Best when many items share one total.

Composition share with a big centre metric — slice labels on the ring, clean legend alongside. Up to 8 slices.

Chronological narrative — events along a horizontal axis, auto-laned so labels never collide, with optional details under each marker.

Process / dependency / system maps — a layered DAG with rounded-rect nodes and bezier-curved edges, laid out automatically.

Bar, line, area, and pie — a single figure, or a grid of small-multiples for a multi-panel dashboard in one scene.

Fallback for novel compositions the editorial primitives don't cover — an AI-authored HTML/CSS/SVG document in a sandboxed iframe.

A Mermaid auto-layout diagram — sequence, ER, state machine, Gantt, or class diagram.

Plain prose with no visual structure — meeting notes, written summaries.

A syntax-highlighted code block, sized for a shared screen.