Tool Execution and Approvals

When agents interact with your connected services, they use tools to read data and take actions. This page explains how tool execution works and how you control what agents can do through permissions and approvals.

What are Tool Calls?

Tools are the actions agents can perform in your connected services. Each tool represents a specific operation, such as:

When you ask an agent to do something, it identifies which tools are needed and executes them to fulfill your request. For example, if you ask "Create a task in Asana called 'Review proposal' due Friday," the agent uses the "Create Task" tool with the appropriate parameters.

Each connection comes with its own set of tools based on what the external service supports. The more connections you add, the more tools your agents can access.

Permission Levels

Every tool has a permission level that determines how the agent can use it. You configure these permissions per tool, per connection.

Allowed

The agent executes the tool automatically without asking for confirmation. Use this for:

Requires Approval

The agent proposes the action and waits for your explicit approval before executing. Use this for:

Denied

The agent cannot use this tool under any circumstances. Use this for:

The Approval Workflow

When an agent needs to use a tool that requires approval, the execution pauses and prompts you for confirmation.

Step 1: Agent Proposes an Action

The agent determines which tool to use and prepares the action. Instead of executing immediately, it presents the proposed action to you in the chat.

Step 2: Review the Proposal

You see exactly what the agent wants to do, including:

This transparency lets you verify the action is correct before it runs.

Step 3: Approve or Deny

You make a decision:

The agent waits for your response and does not proceed until you choose.

Step 4: Agent Continues

After your decision, the agent either:

The conversation continues normally from there.

Approving or Denying Actions

When the agent requests approval, you interact with the approval prompt in the chat interface.

What You See

The approval prompt displays:

How to Approve

How to Deny

Adding Notes

When approving or denying, you can add a note to provide context:

To add a note, type in the note field before clicking Approve or Deny. Notes are optional but useful for record-keeping and clarification.

Configuring Permissions

You set tool permissions in the connection settings. This gives you granular control over each tool in each connection.

Where to Set Permissions

Changes take effect immediately for all agents using that connection.

Best Practices

Reading data does not change anything, so it is generally safe to allow. This lets agents answer questions quickly without interruption.

List Tasks       -> Allowed
Get Project      -> Allowed
Search Contacts  -> Allowed

Creating and updating data has consequences. Requiring approval gives you oversight without blocking functionality entirely.

Create Task      -> Requires Approval
Update Task      -> Requires Approval
Add Comment      -> Requires Approval

Deletion and other irreversible actions should typically be denied. If you need to delete something, do it manually in the original service.

Delete Task      -> Denied
Remove Project   -> Denied
Archive Workspace -> Denied

Adjusting Over Time

Start with stricter permissions and relax them as you build trust:

Viewing Tool History

All tool executions are logged, giving you a complete record of what agents have done.

Where to Find Tool History

What the History Shows

For each tool execution, you can see:

Using History for Auditing

The tool history serves as an audit trail for your team:

Next Steps