The Interact with Page block is a highly versatile tool for controlling the AI agent’s actions on a webpage. You can use it in two ways: by providing a precise, sequential list of commands, or by defining a single, high-level goal that the agent will attempt to achieve by performing multiple sub-steps automatically. This makes it essential for everything from simple button clicks to complex tasks like filling out forms or performing a search.

​
Purpose

Use the Interact with Page block when you need your AI Agent to:
  • Perform Step-by-Step Actions: Click buttons, type text, select from dropdowns, navigate, etc.
  • Achieve a Goal: Execute a multi-step task from a single instruction, like searching for an item or completing a login process, letting the AI handle the intermediate steps.
  • Handle Dynamic Elements: Close pop-ups, interact with menus, or trigger JavaScript events.
  • Control Workflow Execution: Conditionally pause or stop the workflow.

​
Configuration

​
1. Describing the Task

This is the core text area where you define what the agent should do. You can choose one of two approaches:

​
Approach A: Step-by-Step Commands (High Reliability)

For maximum control and reliability, list precise commands, with each new command on a new line. The agent will execute these in the exact order you provide.
  • Example:
    1. type '{{username}}' into 'Username or Email field'
    2. type '{{password}}' into 'Password input'
    3. click on button 'Sign In'

​
Approach B: Goal-Oriented Task (High Flexibility)

You can provide a single, high-level instruction. The agent will then attempt to break this down into the necessary sub-steps. This requires setting a sufficient “Maximum number of steps” in the Advanced Options.
  • Example:
    • Find car part by part number in the search bar. Part number is {{part_number}}.
    • In this case, the agent will attempt to perform a sequence of actions like: closing pop-ups, clicking the search bar, typing the part number, and clicking the search button, all from one instruction.

​
2. Available Commands & Syntax (for Step-by-Step Mode)

  • click: Clicks on a target element (click on button 'Log In').
  • type: Types text into a field (type '{{text}}' into 'Search bar').
  • goto: Opens a specific URL (goto 'https://example.com/contact').
  • press: Presses a keyboard key (press 'Enter').
  • back: Goes to the previous page.
  • idle: Waits for 30 seconds.
  • stop: Immediately stops the workflow run (stop 'Item not found').

​
3. Using Variables

Variables (e.g., {{part_number}}) are crucial for both modes, allowing you to make your interactions dynamic.

​
4. Advanced Options

  • Maximum number of steps: This is critical for goal-oriented tasks. The default is 1. For a goal like finding a car part, you must increase this value (e.g., to 5 or higher) to give the agent enough “action credits” to perform all the necessary sub-steps (close pop-ups, click, type, click again).
creenshot: Interact with Page block configuration showing a step-by-step task

​
Examples

Example 1: Goal-Oriented Search (Recommended for flexibility)
  • Instruction: Find car part by part number in the search bar. Part number is {{part_number}}.
  • Advanced > Maximum number of steps: 5
  • Result: The agent autonomously handles pop-ups, clicks, typing, and submitting the search.
Example 2: Logging into an Account (Step-by-Step for reliability)
  1. type '{{username}}' into 'Username or Email field'
  2. type '{{password}}' into 'Password input'
  3. press 'Enter'
  4. idle
  5. goto 'https://app.example.com/settings/profile'

​
Key Considerations

  • Choose the Right Approach: Use step-by-step commands for precise, linear processes. Use a goal-oriented task for more complex interactions where the exact sequence of clicks might vary or involve handling unexpected elements like pop-ups.
  • Set Step Budget for Goals: When using a goal-oriented instruction, always set an adequate “Maximum number of steps”.
  • Clarity is Key: Whether writing steps or a goal, be as clear and specific as possible.
The Interact with Page block gives you a flexible spectrum of control, from micromanaging every action to defining a high-level goal and letting the AI handle the details.