Skip to main content
Browser Actions execute quick browser interactions using AI (Claude Sonnet 4.5) with a maximum of 2 steps per invocation.

​
Overview

Step type: browser_action Browser Actions are perfect for simple, predictable interactions where you need fast execution and want to benefit from caching.

​
How It Works

  • AI Model: Uses Claude Sonnet 4.5 with the browser-use library
  • Step Limit: Maximum 2 autonomous steps per invocation
  • Variable Support: Supports variable interpolation using {{variable_name}} syntax
  • Caching: Built-in caching for repeated actions
  • Timeout: 300 seconds

​
When to Use

Perfect for:
  • Simple interactions like “Click submit button and verify success”
  • Quick navigation tasks like “Go to pricing page and take screenshot”
  • Single-purpose tasks that don’t need full workflow orchestration
  • Tasks where caching will provide significant cost savings
  • Predictable, repeatable actions
Not suitable for:
  • Multi-step workflows requiring more than 2 actions
  • Complex conditional logic spanning many steps
  • Tasks requiring state management across interruptions
  • Unpredictable flows with many decision points

​
Cost

  • Fresh execution: 50 credits
  • Cached execution: 5 credits
Browser Actions are 10x cheaper when cached! For repeated tasks, the cost savings are significant.

​
Example Use Cases

​
Simple Navigation

"Navigate to google.com and search for 'anthropic'"

​
Form Interaction

"Click the login button and enter email"

​
Quick Verification

"Scroll to footer and click contact link"

​
Screenshot Capture

"Go to the pricing page and take a screenshot"

​
With Variables

"Search for '{{search_term}}' and click the first result"

​
Configuration Example

{
  "type": "browser_action",
  "action": "Navigate to {{website_url}} and click the sign up button",
  "variables": {
    "website_url": "https://example.com"
  }
}

​
Best Practices

​
Do ✅

  • Use for simple, well-defined tasks
  • Leverage caching for repeated operations
  • Keep instructions clear and concise
  • Use variables for dynamic content
  • Combine 2 related actions in a single step

​
Don’t ❌

  • Try to cram complex workflows into 2 steps
  • Use for unpredictable multi-page flows
  • Rely on complex conditional logic
  • Expect state persistence across calls
  • Use when you need more than 2 actions

​
Limitations

  • Max 2 steps per invocation
  • 300 second timeout
  • No state persistence between invocations
  • Cannot handle complex multi-page workflows

​
When to Upgrade

Consider switching to Browser Agents when:
  • You need more than 2 steps
  • The workflow is unpredictable or conditional
  • You need state management
  • The task spans multiple pages with variable paths
Consider switching to Browser Code when:
  • You need precise control over logic
  • You’re extracting structured data
  • You need custom validation or transformation
  • The AI approach is too unpredictable

Browser Agents

For complex multi-step workflows

Browser Code

For precise programmatic control