Audit intake

Tell us what you're seeing.

Four fields, no calendar invite. The agent URL, your vertical, how often you ship, and the one failure mode you most fear. We come back with a written finding list within two business days — the first audit sells itself.

Public · unauthenticated
Replies within 2 business days
Intake form

Four fields.

Inline validation lives on this form; the server re-checks the same schema before persisting. The submission shows up in the operator inbox as a one-line summary, scoped by vertical.

The endpoint we'll point adversarial scenarios at.

Picks the scenario family we run first — refund overrides for travel, KB leakage for marketplaces, etc.

How often the prompt, model, or knowledge base changes.

One paragraph — the failure mode, not the feature request. 500 chars max.

What happens next

A written reply inside two business days.

No sales call before you have something to react to. No automated funnel. Just a short document naming the scenario families we'll run first and where we expect the early P0 / P1s to land.

  • A written finding list

    Within two business days: the scenario families we will run, the rubric tags we will score against, and the rough severity pre-rank. No marketing fluff, no scheduled call before you have something to react to.

  • NDA on request

    Refund-heavy travel and B2B marketplace prospects often need a mutual NDA before sharing the agent URL. We sign yours or ours — whichever moves fastest.

  • Nothing shared without consent

    Your agent URL, transcripts, and finding list stay between us. We never reuse scrubbing scenarios or the failure modes you name against another client without written sign-off.

FAQ

Quick answers.

Why no calendar picker on this page?

Because a four-field written brief produces a more useful first reply than a 30-minute call. The brief is yours to send on your own clock, and ours to respond to in writing.

Will running scenarios against my agent hurt production traffic?

No. Every Picaroon run is sandboxed against a redacted copy of the agent surface or a staging instance you provision. We never point at a live customer-facing endpoint without written sign-off.

What if the failure I most fear turns out to be a feature?

Then the rubric still ranks it. Severity comes from framework tags (OWASP / CSA), not from our opinion. Your team keeps the final say on what ships.