Agents

Agents are AI-powered automations that execute tasks for you. They're the code that runs to accomplish goals like generating weekly sales reports, analyzing customer feedback, or monitoring website metrics. Each agent can run manually, via webhook, or on a schedule, and produces documents and logs.

Creating Agents

Create agents using the chat interface. Describe what you want, and we generate the code for you.

Click New Agent in the app and describe what you want the agent to do. For example: "Generate a weekly sales report from HubSpot" or "Monitor Intercom conversations for negative sentiment". We automatically generate code based on your description.

Understanding Agents

Every agent includes a text description that explains what it does and how it works. This description serves as a self-documenting standard operating procedure (SOP) for the agent, making it easy for you and your team to understand what each agent does without needing to read code.

You can view an agent's description on its page to understand its purpose and functionality. Toggle between the description view and code view using the buttons in the header. The description provides a clear overview, while the code shows the implementation details.

Updating Agents via Chat

You can modify agents anytime using the chat interface. The chat lets you ask questions, request changes, and refine your agents through conversation.

How to Chat

Open an agent's page to see the chat interface on the right side. Type your request in the chat input at the bottom. For example, try "Add error handling" or "Change the report format to include charts". We'll update the agent code based on your request. Keep chatting until the agent does exactly what you need.

You can ask questions like "What does this code do?" to understand how your agent works, or request specific changes to improve functionality.

When you send a chat message, our AI processes your request through several steps to ensure your agent is updated correctly.

  1. Understand your request and make a plan: We analyze what you're asking for and create a plan for how to implement it.
  2. Read documentation of systems you use: We review the documentation for any external systems or APIs your agent uses to ensure we use them correctly.
  3. Re-evaluate plan based on documentation: We refine the plan based on what we learned from the documentation to ensure accuracy.
  4. Implement plan: We update your agent code according to the refined plan.
  5. Double check implementation: We review the code changes to verify they work correctly and match your request.
  6. Update SOP: We update the agent's description (the standard operating procedure) to reflect any changes made.
  7. Reply: We send you a response explaining what we changed and why.

Resetting Chat

If you want to start fresh with the chat conversation, click the New Chat button in the header (the chat icon with a plus). This clears the current chat history and starts a new conversation session for that agent.

Renaming Agents

To change an agent's name, click the menu button (three dots) in the header and select Rename. Enter the new name and save. The agent name appears in lists, breadcrumbs, and throughout the interface.

Organizing Agents

Keep your agents organized by grouping related automations together. The sidebar shows all your agents, and you can drag and drop them to reorder or group them by purpose.

Create groups for different areas of work (like "Marketing", "Sales", or "Support") and move agents into them. This helps you find agents quickly and keeps related automations together. You can rename groups or delete them when they're no longer needed.

Running Agents

Agents can run in three different modes: manually, via webhook, or on a schedule. Choose the mode that best fits how you want your agent to execute. Click the trigger button on the agent page (shows "Run Manually" by default) to change how your agent runs.

Run Manually

Manual mode lets you run agents on demand whenever you need them. Click the Run button (play icon) in the header to execute it immediately. This is useful for testing agents, running one-time reports, or triggering agents when you need them rather than on a fixed schedule.

Run via Webhook

Webhook mode lets external systems trigger your agent in real-time by sending an HTTP request to a unique webhook URL. Click the trigger button and select Run with Webhook to enable webhook mode. You'll see your agent's webhook URL displayed next to the trigger button. Click it to copy the URL. Any system that can make HTTP requests can trigger your agent by calling this URL.

Use webhooks when you want agents to run in response to events from other tools. For example, you can trigger an agent when a new lead is created in your CRM, when a payment is processed, or when a form is submitted.

Run on Schedule

Schedule mode runs your agent automatically at specified times. Click the trigger button and choose from several schedule options. Available schedules include:

For schedules that include a time, a time picker appears at the bottom of the dropdown where you can select when you want the agent to run.

What Happens When an Agent Runs

When an agent runs, you can watch its progress in real time and review the documents it creates once it finishes. The agent creates a log, runs your script step by step, records progress in real-time, creates documents, and completes or reports errors. You can watch agents run in real-time by viewing their logs.

Viewing Logs and Documents

From the agent page header, you can quickly access the agent's execution history and outputs. Two buttons give you direct access:

Deleting Agents

Delete an agent when you no longer need it or want to remove it permanently.

Click the menu button (three dots) in the header and select Delete. Confirm the deletion in the dialog that appears. Deleting an agent permanently removes it and all associated logs and documents. This action cannot be undone.