Security

The page to forward to whoever has to say no

Written for your security reviewer, not for the person who wants to buy. It sets out where the AI coworker runs, what it can reach, what it cannot, and what we have not earned yet.

Get a demo
What a coworker can and cannot reachInside the granted scope it reaches the systems your team named. Outside it, nothing. Access arrives as an IAM role you issue and leaves as a line in your own audit log.CAN REACHThe CRMscope you grantedThe mailboxone shared inboxThe ERPread, and postInternal databasebehind your VPNIAM roleyou issue itAudit logyou keep itCANNOT REACHAnything your team did not grantSystems outside the job it was built forThe open internet, unless your egress rules allow it

Most AI vendors spend the security review explaining how they will protect your data once it reaches them. We do not have that conversation, because your data does not go anywhere.

WHAT STAYS YOURS

Every control here is one you already operate.

Nothing asks you to trust a boundary you cannot inspect.

Runs in
YOUR CLOUD
Access via
YOUR IAM ROLES
Logged to
YOUR AUDIT TRAIL
Bounded by
YOUR EGRESS RULES

Where it runs, what it reaches, who approves

The eight answers a security reviewer asks for, before they ask for them.

Where it runs

A service inside a VPC in your own account. Not a tenant in a platform we operate.

How it authenticates

IAM roles your team issues, scoped per system, with no standing credential held by us.

What it can reach

Only the systems you connect, at the narrowest permission the job actually needs.

Who approves what

Consequential actions stop and wait for a person. You draw that line, per action type.

What gets logged

Every action, every escalation, and every refusal — written to the audit trail you collect.

Egress

Outbound traffic follows your network policy. What may leave your perimeter is your configuration.

Data residency

Your account, your region. There is no vendor-side copy of anything to locate or delete.

Turning it off

Revoke the roles and access ends. Stop the service and it is gone, without involving us.

What it did, in a log you already read

There is no vendor-side console to request access to. Everything the coworker does is written to the audit trail your own stack already collects, including the actions it declined to take alone.

See how access works   →
Access logyour audit trail
RL
Assumed crm-read-onlyIssued by your team · expires in 4h
Recorded
WR
Wrote to ticket #4812Scoped write · helpdesk only
Recorded
RF
Refund above policy limitStopped and asked a person
Held
EG
Outbound call blockedDestination outside your egress rules
Denied

Questions your reviewer will ask

Answered the way we would answer them on a call, including the ones where the answer is no.

Where is our data processed?
Inside your own cloud account. The AI coworker is deployed into a VPC you control, so the records it reads and writes never leave the environment they already live in. There is no vendor database holding a copy of your business data.
What credentials do you hold?
None that persist on our side. The coworker assumes IAM roles your team creates inside your account, scoped per system to the narrowest permission the job needs. Revoking a role ends that access immediately, without our involvement.
Can it act without a human approving?
Only for the actions you designate as routine. Anything you mark consequential, such as sending an external email, changing a record, or releasing a payment, stops and waits for a person. You set that line during deployment and can move it at any time.
What can your team see?
Whatever you grant, and nothing more. Support access is issued the same way any vendor's is in your cloud: a role you create, scoped, time-boxed, and logged in your own audit trail. You can withdraw it without breaking the deployment.
Which certifications do you hold?
None yet, and we will not display a badge we have not earned. We are pre-launch and working with design partners. What we can do instead is walk your reviewer through the architecture line by line, because the security argument here is structural rather than procedural.
How is the model provider handled?
You choose it, and you control the egress path out of your own network. Because the coworker runs inside your account, the rules governing what it may reach beyond your perimeter are configured by your team rather than assumed by us.
What happens if we terminate?
You revoke the roles and tear down the deployment in your own account. Nothing has to be deleted on our side, because nothing of yours was ever stored there. The work the coworker did stays where it was written: in your systems of record.
“We hold no certifications yet, and we will not show a badge we have not earned. What we can do is walk your reviewer through the architecture line by line.”
— Where we are, stated plainly
Access logsource: your audit trail

Role assumed: crm-read-only

expires in 4h · issued by your team✓ Recorded in your stack

Put an AI coworker
inside your own cloud.

Bring your security reviewer to the first call. That is the conversation we want.

Get a demo