Admin Settings → Agentic governs the autonomous AI capabilities covered in Agentic Workflows — agents that plan and act with tools rather than following a fixed script.
What you control
| Control | Effect |
|---|---|
| Agentic capability enablement | Whether agents (including the Deep Agent workflow node) are available at all, via the feature toggle and Permissions. |
| Agentic Categories | Grouped catalogues that organise which agents/tools each audience can reach. Create a category, name it, and assign it to a group; users see the categories their group has. Duplicate names within a group are rejected. |
Managing categories
| Action | Notes |
|---|---|
| Add Category | Create a named category (for example Finance agents, Lab tools). |
| Rename | Categories rename in place; users see the new name. |
| Delete | Removes the category — confirm intentionally; assignments relying on it change visibility. |
Governance model for autonomous capability
Agents combine three powers — reasoning, tools, and persistence. Govern all three:
- Enable narrowly — agentic features for vetted groups first; broaden with evidence.
- Tool scoping — agents are only as risky as the tools they can reach; keep production agents read-only where possible.
- Human gates — encourage User Approve nodes before side effects (built into the workflow engine).
- Audit trails — every agent run is recorded in Workflow History; know where to look when reviewing incidents.
- Guardrails — apply input/output policy checks. See Guardrails.