Trust & Security

Built so the AI can't go rogue with your money.

You're letting software talk to your customers and move money. Here's exactly what stops it from doing the wrong thing — enforced in code, not left to a prompt.

AI safety, mechanically enforced

Five rules the agent can't talk its way around.

01

Read-back confirmation

Every tool whose effect includes a charge, a cancellation or a date change refuses to run until it's called with the customer's confirmation. Without it, the agent is told to read the specifics back and get a yes.

02

Completion-claim guard

A reply that says a booking, cancellation, reschedule, swap, pickup, payment or credit went through — without the matching tool having actually succeeded on that call — is blocked before it reaches the customer, and the conversation is handed to a person.

03

Idempotent writes

Every write requires an idempotency key. A retried call with the same key returns the first result instead of doing the work twice.

04

Prices come from your rate card

The agent never calculates a price. Quotes, extensions and balances come from the same pricing engine as your console and booking widget.

05

Credits need a human

The agent can submit a capped service credit for review. It can never apply one, and it is never allowed to tell a customer that it has.

Payments

Card data stays out of our hands. And out of the AI's ears.

Stripe-hosted card entry

Cards are entered into Stripe's own fields and payment links. Card numbers don't pass through CansOut servers, keeping us in PCI SAQ-A scope.

Keypad capture on calls

For phone payments the call is handed to the carrier's secure keypad capture. While it's in progress, the audio the AI listens to carries nothing — digits are never transcribed.

No card numbers by voice

The agent is instructed never to ask for a card number out loud, and the platform has no way to accept one that way.

Platform

The boring, essential stuff.

Tenant isolation

Every record belongs to exactly one operator, enforced by a global query filter. A request without a tenant context fails instead of guessing.

Separate sign-in realms

Operators, drivers and customers sign in separately. Operator accounts support authenticator-app MFA; drivers use per-device PINs; customers use one-time e-mail links.

Role-by-permission matrix

Grant exactly what each role needs across the back office — not a handful of fixed roles.

Encrypted in transit

HTTPS everywhere with HSTS. Stripe, Twilio and SendGrid callbacks are verified by signature before anything is processed.

Append-only event log

Every change records who made it — user, driver, customer, AI agent, system or integration.

Messaging consent

SMS consent is tracked for TCPA and 10DLC rules, so texts only go to people who've agreed to receive them.

Our engineering baseline is mapped against OWASP ASVS 4.0.3 Level 2, with automated gates for static analysis, dependency advisories, secret scanning and authorization conformance. Security questions or a responsible-disclosure report: [email protected].

Ask us anything

Bring your hardest question to the demo.