Agentic AI › Build
Tool Grants
Which agent may call what, and on what terms. The same tool is a different risk in two agents' hands, which is why the limits live on the grant rather than on the tool.
Where to find it
- Module: Agentic AI
- Group: Build & Orchestrate
- Section: Build
- Screen: Tool Grants
Address in the console: /agentic/agent-tools
List · 14 columns · 8 fields on its form · 2 of them required
What the list shows
These are the columns of the list, in the order they are drawn. Any of them can be sorted on, filtered on, hidden, or exported.
| Column | Kind | Field name |
|---|---|---|
| Agent | Text | agent_name |
| Tool | Text | tool_name |
| Kind | Status | tool_kind |
| Changes data | Yes or no | is_write |
| Moves money | Yes or no | moves_money |
| Personal data | Yes or no | touches_pii |
| Reaches out | Yes or no | is_external |
| Always ask a person | Yes or no | requires_approval |
| Calls (30d) | Number | calls_30d |
| Enabled | Yes or no | is_enabled |
| Fixed arguments | Text | bound_args |
| Only when | Text | precondition |
| Calls per run | Number | max_calls_per_run |
| Since publish | Text | since_publish |
The form
What is asked for when a record is created here. Required fields are marked; a choice drawn from a configuration list is one an administrator can extend without a release.
| Field | Type | Choices | Notes |
|---|---|---|---|
| Agent required | Choice | Looked up from existing records | |
| Tool required | Choice | Looked up from existing records | |
| Fixed arguments | Text | Values this agent must always use with this tool, written as JSON: names and values in braces, like the example in the box. They replace anything the agent chooses for the same name, so {"business_unit_id": "BU-01"} keeps it working in that business unit only. Leave empty for none. | |
| Only when | Long text | A condition every call must meet, written on what the call sends: the name of one of the tool's inputs, a comparison (=, <>, <, <=, >, >=) and a value, joined with and / or -- for example amount <= 5000, or status = 'draft'. A call that does not meet it, or does not send what it names, is refused and the agent is told why. Checked when you save. Leave empty to allow every call. | |
| Calls per run | Number | The most times the agent may use this tool in one run, as a whole number. Further calls are refused and counted as Refused calls on the run. Empty: no limit of its own (the agent's own tool-call limit still applies). 0: the tool cannot be called at all. | |
| Organizational unit | Choice | Looked up from existing records | Which business unit or department this record belongs to. Defaults to the unit selected in the top bar. |
| Always ask a person | Yes or no | If ticked, every call to this tool waits until a person approves it. The request appears in Agentic AI › Agent Approvals, where anyone whose role may act on Agent Approvals approves or declines it; the agent's run waits until then. | |
| Enabled | Yes or no | Untick to withdraw the grant without deleting it. It stops at once, without a Publish. |
The rules that apply here
Everything on this screen obeys the platform rules rather than rules of its own. The ones worth knowing before you use it:
- Narrowing a list — search, conditions, sort, totals, export and saved views.
- What may be changed — the five grounds a record may not be edited on.
- Companies and business units — what the switcher above this list is doing to it.
- Creating a record — required fields, where choices come from, and how the number is allocated.
- Importing a spreadsheet — this screen accepts one, and it runs through the same rules as typing.