Documentation

The eight tools

Every simulation starts in the PromptBar. Pick one of these tools, type your question, send:

• Ask — stance distribution + representative quotes on a question. The signature tool.
• Compare — A/B (or A/B/n) test 2–6 framings against the same cohort. Returns a gauge + per-framing quote.
• Segments — Ask scoped to a niche cohort ("Turkish-German women under 35") with segment breakdown.
• Panel — 10 agents debate the topic over 30 turns; ends with a synthesis paragraph.
• Asset — upload an image (poster, ad, package design); get an attention heatmap plus per-demographic reactions.
• Report — export the current run as a shareable PDF via the browser’s native print-to-PDF.
• Cascade — information-cascade / spiral-of-silence simulation: does this claim spread, silence itself, or polarize when exposed to a cohort, wave by wave?
• Counterfactual — same cohort under two conditions (baseline vs perturbation) → stance delta.

The data model

Each tool runs against synthetic personas: statistical profiles grounded in three data sources:

• Public data — census, open government statistics, published polling, electoral records.
• Licensed datasets — commercial data we purchase (Kantar and others).
• Client-proprietary data — your organization’s own data, scoped to your tenant.

No real person’s personal information is ever used. See the Responsible AI page for details.

Refining a run

After a run completes you can:

• Filter the cohort down with a yes/no question ("Kreuzberg renters only"). The card recomputes against the kept subset; quotes are regenerated from kept-subset representatives.
• Replay the exact same cohort with a different prompt — isolates the effect of the wording change.
• Export any run as a PDF.
• Rate quotes and results with thumbs up/down; comments optional on thumbs-down.

Credits and billing

Every user starts with a $15/day soft cap. A typical Ask run costs ~$1–2; Compare and Panel run higher. When you hit the cap, further runs return 402 Payment Required until midnight UTC; monthly caps apply alongside. Admins can raise per-seat or per-org overrides.

Admin panel

Users with the admin or owner role see a /admin route with four tabs:

• Feedback — every thumbs-up/down your org has left, filterable by rating.
• Users — seat list with credit usage.
• Runs — recent runs, filterable by status.
• Audit — who did what when; filterable by action prefix (run., admin.,feedback., account.).

API access

A public API with per-seat keys ships next. Early access: contact hello@kukilab.tech.

Your data, your control

• Export everything we hold about you from Settings → Account → Export my data.
• Delete your account from Settings → Account → Delete my account. See the Privacy policy for exactly what is anonymized versus retained.

Need help?

Reach us at hello@kukilab.tech. Live status: /status.