Skip to main content
Agent Builder is in private preview. Sign up for the waitlist today.
Agent Builder lets you turn natural-language ideas into production agents. It’s powered by deep-agents, and is not .

​
Triggers

Triggers define when your agent should start running. You can connect your agent to external tools or time-based schedules, letting it respond automatically to messages, emails, or recurring events. The following examples show some of the apps you can use to trigger your agent:

Slack

Activate your agent when messages are received in specific Slack channels.

Gmail

Trigger your agent when emails are received.

Cron schedules

Run your agent on a time-based schedule for recurring tasks.

​
Sub-agents

Agent Builder lets you create sub-agents within a main agent. Sub-agents are smaller, specialized agents that handle specific parts of a larger task. They can operate with their own tools, permissions, or goals while coordinating with the main agent. Using sub-agents makes it easier to build complex systems by dividing work into focused, reusable components. This modular approach helps keep your agents organized, scalable, and easier to maintain. Below are a few ways sub-agents can be used in your projects:
  • Handle distinct parts of a broader workflow (for example, data retrieval, summarization, or formatting).
  • Use different tools or context windows for specialized tasks.
  • Run independently but report results back to the main agent.

​
Human in the loop

Human-in-the-loop functionality allows you to review and approve agent actions before they execute, giving you control over critical decisions.

​
Enabling interrupts

1

Select a tool

When configuring your agent in Agent Builder, select the tool you want to add human oversight to.
2

Enable interrupts

Look for the interrupt option when selecting the tool and toggle it on.
3

Agent pauses for approval

The agent will pause and wait for human approval before executing that tool.

​
Actions on interrupts

When your agent reaches an interrupt point, you can take one of three actions:

Accept

Approve the agent’s proposed action and allow it to proceed as planned.

Edit

Modify the agent’s message or parameters before allowing it to continue.

Send feedback

Provide feedback to the agent.
Edit the source of this page on GitHub.
Connect these docs programmatically to Claude, VSCode, and more via MCP for real-time answers.
⌘I