ATTENTION API · REST + MCP

AI can generate 50 pages an hour. Something has to judge them.

The eval layer for AI-generated UIs: a deterministic attention model behind one endpoint. Your agent generates variants, the API scores where a human eye will land, the best one ships. ~10 seconds per score.

eval-loop.ts
// The eval loop — generate N, score all, ship the winner
const variants = await agent.generate(brief, { n: 5 });

const scored = await Promise.all(variants.map(async (v) => {
  const r = await fetch("https://heatpoints.com/api/v1/analyze", {
    method: "POST",
    headers: { "X-API-Key": process.env.HEATPOINTS_API_KEY,
               "Content-Type": "application/json" },
    body: JSON.stringify({ url: v.previewUrl }),
  }).then((r) => r.json());
  return { ...v, score: r.score, hierarchy: r.metrics.hierarchy };
}));

const winner = scored.sort((a, b) => b.score - a.score)[0];
await agent.deploy(winner);   // ship eyes-first, not vibes-first
RUN OUTPUT — 5 VARIANTS RANKED

variant-c ships

hero contrast + single CTA

74

variant-a

original brief

61

variant-e

video background

55

variant-b

3-column hero

48

variant-d

gradient everything

37

Deterministic

Same pixels, same score — no LLM judgment drift. Deltas between runs measure YOUR changes, nothing else. That's what makes it usable in CI and agent loops.

Grounded in eye-tracking science

The model (UNISAL) is peer-reviewed and trained on real human gaze datasets — the score approximates a first-time visitor's eye, not an aesthetic opinion.

Fast & cheap enough to loop

~10s per URL, milliseconds per uploaded image. Score every variant, every PR, every deploy — not just the final one.

One endpoint, two inputs

POST /api/v1/analyze { url }

Any public URL — production, staging, preview deployments. Returns score /100, attention intensity, focus clarity, hierarchy, coverage, primary gaze zone; optional composited heatmap (include_heatmap).

POST /api/v1/analyze { image_b64 }

A screenshot, Figma export, thumbnail or ad creative — score designs that aren't deployed anywhere. Same output.

PRICING — CREDITS INCLUDED IN EVERY PLAN

Free

0 €

10 analyses/mo

Pro

29 €

300 analyses/mo

Agency

99 €

1000+ analyses/mo · white-label

1 credit = 1 analysis, URL or image. Need serious volume for an agent product? Talk to us — volume packs are coming.

Stop shipping the first draft your agent liked.

Free account = API key + 10 analyses/month. First score in 2 minutes.

Get your API key →