Skip to main content
An Action block defines a follow-up task that runs after an Agent block produces its output. Always emailing the Agent’s response is an example of an Action. You can connect several Action blocks to one Agent block’s output. For example, to email the same response, record it in Notion, and notify Slack, wire the Agent block’s output to each Action block.

​
Tool and Action differences

A Tool is something the Agent chooses and uses as needed while producing an answer. An Action, by contrast, runs according to the Flow’s wiring, so the Agent does not decide whether to run it. Use a Tool when the Agent needs to decide during reasoning whether a task should run. Use an Action when a task must follow from the Agent’s decision.

​
Action settings

Like a Tool, an Action uses one of three runtimes: HTTP, Script, or Connector. See the Tool block for how to configure each runtime. Assign an icon to an Action block to distinguish Action blocks on the canvas. To use no icon, click the [Remove] button.

​
Approval

For high-risk Action blocks, such as processing refunds, deleting database data, or sending notifications, turn on Require approval. Approval is required before the Action runs, and execution waits until it is granted.

​
Retry

An Action has no retry settings of its own. Configure failure handling, such as timeouts, circuit breakers, attempt and cost budgets, and what happens when they are exhausted, in the Retry tab of the connected Agent block before it.