I·K·I·G·A·ISpeak · Ship · Scale
Enterprise
Most platforms ask you to trust a badge. This page names each control, says what enforces it, and says plainly where a control is implemented but not yet certified. Tenant isolation is a database rule rather than a promise from the application layer, provider keys are envelope-encrypted and can be your own, and every schema change is recorded in a ledger you can read.
The claim every buyer tests first: can one customer reach another customer's data. The boundary is in the database, so it holds even when the application layer is wrong.
Ninety-nine tables carry a tenant key and every one of them has RLS enabled. Isolation is a policy in Postgres, not a WHERE clause someone has to remember to write.
Twenty-seven internal tables have RLS on and no policy at all, which denies every request that is not the service role. A table added without a policy is closed, not open.
Your workspace memory is scoped to your organisation. The shared learning layer that improves the agents refuses any row carrying a user, an organisation, a stored conversation or contact details. See the published disclosure.
A weekly job creates two synthetic organisations and attempts cross-tenant reads and writes through the live API. Every probe must block. A single leak is treated as a stop-everything event.
Who can sign in, what they can reach once they are in, and what is written down when they act.
Implemented and enabled for organisations that require it, with a domain policy that can force SSO for everyone on your domain. It is off by default rather than half-on.
Membership is explicit and role-scoped. Read access, write access and administrative actions are separate, and the boundary is checked server-side on every privileged route.
Credential changes, price changes, sending-identity changes and administrative actions are recorded with the actor. The trail is append-only and is not reachable from a browser session.
Inactivity locking, a PIN gate on sensitive surfaces, and scoped tokens for anything that runs outside the browser.
What we hold, how it is protected, and how you get it back or have it destroyed.
Every model slot accepts your own provider credential. Keys are envelope-encrypted at rest and the plaintext never reaches a log. With BYOK in force the call bills to your account and the conversation never touches a managed key.
Encrypted at rest and in transit on every hop. Stored credentials are wrapped with a master key that is rotatable without re-collecting anything from you.
Account export and deletion are self-serve. Erasure cascades across the tables that hold personal data and writes a compliance record that survives it. Billing records are anonymised rather than destroyed, because tax law requires them kept.
The full register is public and versioned, so you can diff it rather than ask. Read the register.
What is in force today, and what is mapped but not yet certified. The second list matters more than the first when you are the one signing.
How a change reaches your workspace, and what stops a bad one. This is the part of a platform that decides whether an incident is an afternoon or a quarter.
Schema changes run through a ledger that records what was applied and when. A migration cannot be applied twice and an applied one cannot go missing from the record.
Each migration is rehearsed against a pre-production schema in the same engine before it is allowed near your data. A rehearsal failure stops the change.
A build-time guard validates every query the application makes against a committed snapshot of the real schema. Asking for a column that does not exist fails the build rather than a customer request.
Agent behaviour is scored against fixtures on a fixed sampling seed, so a change in the score means a change in the code rather than a roll of the dice.
The order that wastes the least of your time.
Unlimited credits, dedicated workspaces, BYOK required, negotiated terms. Questionnaires and DPA answered before a call, not after one.