Approval workflows
BetaA multi-step approval chain on a contract, routed by role and value — plus a review gate on AI output.
- ✓Ordered approval steps with a required role and a minimum approval count per step
- ✓Value-conditioned routing on thresholds you set — none is a default
- ✓Decision, decider, timestamp and note recorded on every step
- ✓Cancel a request in flight
What it is.
Deal Desk is the real approval feature. A policy defines an ordered ladder of steps, each requiring a role and a minimum number of approvals, with conditions on contract value that you set — no default ladder is seeded on this product, so no money threshold is assumed for you. A contract raises an approval request, steps activate in order, and each decision records who made it, when, and their note.
Separately, AI review is a gate at the other end: hold a class of AI output above a risk threshold until a human approves it, with deciding authority tied to the permission matrix. It is off by default and is the control to reach for when AI work product must not leave the building unreviewed.
One thing to know: an approval step can carry a deadline and nothing enforces it. There is no auto-approve, no auto-reject and no escalation when a step goes past its date — the deadline is shown, not acted on. And approval gates inside an agent workflow do not exist; that step type is declared and unimplemented.
Three steps.
End to end.
Ordered steps, each with a required role, a minimum number of approvals, and a value condition you choose.
A contract enters the chain. Steps activate in order; each carries a deadline if the policy sets one.
Approve or reject with a note. Who decided, when and why is recorded. Cancel the request to stop it.
What you get.
- ✓Ordered approval steps with a required role and a minimum approval count per step
- ✓Value-conditioned routing on thresholds you set — none is a default
- ✓Decision, decider, timestamp and note recorded on every step
- ✓Cancel a request in flight
- ✓AI review — hold AI output of a given kind above a risk floor for human approval, off by default
- ✓Edit proposals give a single-reviewer approve, reject or withdraw gate on document changes
Quick answers.
Nothing automatic. The deadline is stamped and displayed; there is no job that auto-approves, auto-rejects or escalates.
No. A human-approval step type exists in the agent schema and the engine does not implement it — a run that reaches one fails. Approvals live on contracts and on AI output, not inside agent runs.
Yes. Every decision carries a note, and it is part of the record.