Governance
Trust AI because you control it.
Who may do what, who approves, and what happens when it is wrong: written as rules the system runs.
Scroll through each step
Where a customer message stops.
An agent drafts
With a source behind every number.
Policy gates
Off-policy pricing, the spend ceiling, who may read what.
Off-policy is flagged.
A person approves
Someone reads the draft, then approves or returns it.
Checked at the API and again in the database on write. No setting turns it off.
Send and record
Who approved, when, and the data as it stood.
What it leaves
A complete trail for an auditor.
Who approves what.
Real permissions live in workspace grants; roles only change what the interface puts first.
TaskDraft a quote
Who canAn operator
Where it stopsOff-policy pricing is flagged
TaskApprove and send to a customer
Who canSomeone with approval rights
Where it stopsThe API and the database
TaskChange the pricing policy
Who canA manager
Where it stopsVersion and author recorded
TaskSee raw financial data
Who canAccount owner and admin
Where it stopsChecked in the data layer
Four constraints we don't negotiate.
Not approved, not sent
Checked at the API and again on write.
A workspace is a real boundary
Enforced in the data layer, not hidden in the interface.
Every change leaves a mark
Adjustments create new records; history is not overwritten.
States are never faked
Live, demo, error, waiting and blocked show as they are.
Common questions.
What if we want the AI to send on its own?
We do not offer that. Approval is a runtime constraint, not a configuration option.
Is our data used for training?
No. It stays in your workspace and is not used to train models.
What do we show an auditor?
Every run has timestamps, the accountable person and the data as it stood. From a report line back to the source document.
Show this to your compliance lead.
We would rather take the hard questions now than have the problem surface in month three.