Skip to main content

Set up tool confirmation

Tool confirmation lets you add a human approval gate before an agent executes a sensitive or irreversible tool call, for example, writing a record, sending a message, or triggering an external API. When a covered tool is invoked, Creatio AI Studio pauses the agent and routes an approval request to the appropriate reviewer. The tool runs only after the request is approved. If it is denied or times out, the agent receives a refusal and continues without executing the tool.

You can create a tool confirmation policy at the following levels:

  • Global — applies to every agent in the workspace. Create it from Trust & Governance → Policies section.
  • Agent — applies to one specific agent. Create it from the agent's Govern tab or from Trust & Governance → Policies section.

Before adding an agent-scoped policy, check the Global policies applied to this agent block on the agent's Govern → Policies page to see which global policies are already in effect.

Create a global tool confirmation policy​

  1. Go to Trust & Governance → Policies section.
  2. Click New in the top right. This opens a window.
  3. Select Tool confirmation. This opens a page.
  4. Fill in the policy settings.
  5. Click Create policy.

As a result, the policy will appear in the Policies list and will be applied to every agent in the workspace according to its enforcement mode.

Create a per-agent tool confirmation policy​

  1. Open the agent from the Managed Agents → Agents section.
  2. Click Govern in the top navigation bar.
  3. Go to the Policies tab → Add agent policy.
  4. Select Tool confirmation.
  5. Fill in the policy settings.
  6. Click Create policy.

If you have an existing agent-scoped policy you want to reuse, click Attach policy instead of Add agent policy.

As a result, the policy will appear on the agent's Policies tab and apply only to that agent.

Policy setting reference​

Tools​

In the Tools section, click Add tool and select every tool whose calls must require approval. A single policy can cover multiple tools.

Approval & confirmation​

This section controls who receives the approval request and how long the policy waits for a response.

Routing determines who approves the action:

Option

Description

Self-confirmation

The user who is actively chatting with the agent approves inline. If no response arrives within the in-session timeout, the request falls back to delegated approval automatically.

Delegated approval

The request is routed to a designated approver via an approval channel. The active user does not see the inline prompt.

Self-confirmation fields:

  • In-session timeout (minutes) — how long to wait for the active user to respond before falling back to delegated approval. Out of the box, 5.
  • Confirmation message — the message shown to the user. Leave empty to use the default template. Supported variables: {tool name}, {changes details short}, {changes details full}.

Delegated approval fields:

  • Approval channel timeout (minutes) — how long to wait for the approver to respond before the agent receives a refusal. Out of the box, 60.
  • Approver role — the role whose channel routings receive the approval request. Leave empty to use the legacy /approvals queue.
  • Pending notification (to the requester) (optional) — a message sent to the requester in synchronous channels while approval is pending. Not sent in asynchronous channels such as email. Supported variables: {tool_name}, {policy_name}, {approver_role}, {timeout_minutes}. Maximum 500 characters.

General settings & runtime​

Field

Description

Policy name

A human-readable name to identify the policy in the Policies list and in notifications. Required.

Scope

"Global" — applies to all agents.

"Agent" — applies to the agent to which it is attached.

Enforcement

"Disabled" — the policy is created but does not intercept calls.

"Observe" — intercepts and logs calls without blocking (default).

"Enforce" — blocks tool execution until approved.

Severity

The risk level assigned to this policy: "Low", "Medium" (default), "High", or "Critical". Used for filtering and reporting in the Governance section.

Choose an enforcement mode​

Start with "Observe" to confirm that the policy matches the intended tools without disrupting the agent. Review intercepted calls in the Governance section. Switch to "Enforce" once you are confident the scope is correct and want real approvals to take effect. Use "Disabled" to deactivate a policy temporarily without deleting it.

See also​

Creatio AI Studio overview