Thousand Words

Thousand Words
*A picture is worth a thousand words.*

Turn a sentence into an interactive visual. You describe what you want in plain
language; an LLM emits a tiny JSON spec; a bundled vanilla-JS runtime renders it to a
standalone HTML file (or a static SVG) you open in any browser. Once rendered,
dragging a slider, stepping through an algorithm, or editing a diagram costs zero model
calls — it's all local.

This one README is both the quick reference and the full manual. Start with Quick start,
then the sections below explain the problem, the solution, real use cases (with images),
the everyday workflow, and what lives in the repo. Every field-level detail is in
§ Reference.

Quick start

Install the Claude Code plugin — one marketplace add, one install:

That's it. The install auto-registers both MCP servers (render_gui and
render_diagram) — nothing else to configure. Now just ask an LLM in plain English; it
composes the spec, calls the tool, and hands you a standalone .html.
*"a take-home pay calculator — my numbers stay local"* · *"walk me through bubble
sort"* · *"flowchart of our login flow, grouped by kind"*

Updating to a new release is the same two steps — refresh the
marketplace, then the plugin:

Then start a new session (or /clear) — the SessionStart hook that routes
visual requests to the skills fires at session start, so the update takes
effect from the next session on.

Current release: 0.4.0 — select options can be {label, value} objects,
blank table cells are treated as missing instead of deflating averages to
zero, tables support per-row derived computed columns reachable by
aggregates and CSV export, fsm state machines honor a top-level accept
array, and diagrams scale to fit their viewport instead of clipping wide
graphs. Full history in…