Security and control
AI that works with guardrails
The point is not that the AI is careful. It is that the most damaging things are not available to it — decided outside the model, from your configuration, every single time.
Nothing reaches a customer without approval
Importing orders, connecting a shop or receiving a webhook never contacts anybody. Somebody at your end presses go, every time.
The AI cannot be talked into a job you left off
A job you have not switched on is never offered to the AI at all. It is not attempted and refused — it is not there.
Every handoff ends somewhere
A set of escalation rules whose last step could leave a customer with nothing is refused before it can be switched on. Not handled hopefully at runtime — refused.
You can see why, not just what
When work goes to one person and not another, every candidate and every reason is recorded. So is every action the AI took, against the customer it took it for.
Your data is separated at the database, not by convention
Every record belongs to exactly one business and the database enforces it. Cross-business access is tested as a matrix over every kind of record, and an untested one fails the build.
Credentials are stored as references, never shown
No screen in the product renders a secret, because no screen has one to render. A provider that cannot prove a webhook came from it is refused, and you can see the refusals.
We claim no certification — not SOC 2, not ISO 27001, not any other — because none has been undertaken. What is above is what the software does, and you can check every line of it inside the product.