Getting started
Provenance
In regulated products, a user must always be able to tell what a person wrote from what an AI generated. agentic makes that a default: every AI-generated element carries the same accent and label — one consistent signal across the whole kit.
One accent, one meaning
Provenance isn't a component — it's a pattern. A single design token, --agt-ai-color (and its gradient --agt-ai-gradient), marks everything the AI touched: the streaming cursor, the "Generated by AI" label, the confidence badge. Restyle that one token and the AI signal updates everywhere. It stays deliberately distinct from your primary / brand colour, so "AI" never blends into "interactive".
In a summary
An AI-written summary is labelled and can be regenerated — never mistaken for a human note.
Low risk. The passport was extracted with 98% confidence and the name matches the application exactly. No sanctions or PEP hits.
Recommendation: approve with a standard 12-month review.
In a form field
A suggestion appears as ghost text with a confidence badge. Accept it and the value is marked AI; edit it and it flips to "edited by human" — the provenance follows the value, which is exactly what an auditor needs.
In chat
The agent's replies stream under the same accent, with a collapsible reasoning trace and a live cursor — it's always clear which turns came from the model.
Ask the KYC agent
Why it matters
"The AI did it" is not an acceptable answer in fintech, KYC or legal workflows. Clear provenance is what lets a reviewer trust — and take responsibility for — an agent's output. agentic gives you the UI building blocks; compliance itself is your product's job.