Tool approvals
Add a confirmation gate before sensitive tools run, so a person reviews and approves the agent's proposed action before it executes.
Some actions are too sensitive or irreversible to run without a person's say-so. Tool approvals add a confirmation gate: the agent proposes the action, and someone approves or rejects it before it runs.
How approval works
The agent proposes an action
When the agent decides to call a tool that requires approval, it pauses instead of running it.
An approval card appears
The chat shows a card with the tool name and the exact inputs the agent wants to use.
You approve or reject
- Approve — the tool runs and the agent continues with the result.
- Reject — the tool is skipped and the agent continues without it.
Action states
While an agent is working, each tool call shows its current state:
| State | Meaning |
|---|---|
| Running | The tool is executing |
| Awaiting approval | Paused, waiting for someone to confirm |
| Completed | Finished successfully |
| Rejected | A person declined the action |
| Error | The tool failed to run |
Approvals and autonomous runs
Approvals are especially useful with triggers, where the agent runs on its own. A trigger can do all the analysis automatically and still stop for a human before it sends a reply or changes data.
Related: workflow approvals
Automation workflows have their own approval step. See Workflow approvals to build human-in-the-loop automations.
Versions & publishing
Edit an agent safely as a draft, then publish numbered, immutable versions that control exactly what your production agents run.
Agent templates
Start from a pre-built template instead of a blank agent to get a tested starting prompt and sensible defaults for common support roles.