Keep people in control
Put people in front of risky actions
Reserve human review for the actions where judgment or policy requires it.
Gate any tool you choose - a write, or a read whose query is the risk; everything else continues without a pause.
What this changes for your team.
Agent authors pick, per tool, which of an agent's tools pause for a person - writes, and reads too, each badged so the cost of gating a read is visible. In a conversation the person who asked is given the call to confirm; unattended, or on someone else's credentials, the call waits in Approvals for whoever may decide it. Tools that are not selected proceed normally, so the policy is explicit rather than an implied blanket safety promise. An organization can also require approval on a tool for every agent at once, and no agent can remove that.
How it works in practice.
- 01
Review every bound tool, each badged as a read or a write.
- 02
Select the tools that must stop for a decision, and optionally a condition under which a call may skip the pause.
- 03
Publish the agent so the approval policy is versioned with its setup.
What you can plan around.
What to expect, including limits that matter to your team.
Approval policy stores concrete tool identifiers rather than matching generated call text.
The policy is checked before a gated call is dispatched; an organization-required tool pauses whatever the agent's own setting says.
An ungated call is audited but does not acquire an approval step automatically.
Bring one real process
See how Yekar.AI fits the way you work.
Start with a job your team already owns, plus the tools and decisions around it.
Get started