Web Automation Agents

Agent Types

LiteAgent supports multiple web automation agents, each with unique characteristics and approaches to web interaction.

Supported Agents Comparison

Agent	Type	Vision	API Required	Dark Pattern Detection	Performance
BrowserUse	AI-Powered	✅	OpenAI/Anthropic	High	Fast
DoBrowser	Extension	✅	DoBrowser API	Medium	Medium
MultiOn	Multi-modal	✅	MultiOn API	High	Fast
Agent E	Enterprise	✅	Anthropic	Very High	Medium
Skyvern	Computer Vision	✅	Skyvern API	High	Slow
WebArena	Research	❌	OpenAI	Low	Fast
VisualWebArena	Research	✅	OpenAI	Medium	Medium
Human	Manual	✅	None	Baseline	Variable

Agent Architecture

All agents inherit from the WebAutomationBase class, ensuring consistent behavior:

class WebAutomationBase:
    """Base class for all web automation agents"""

    async def __call__(self):
        """Main execution method"""
        await self.setup_browser()
        await self.navigate_to_site()
        await self.run_task()
        await self.cleanup()

    async def run_task(self):
        """Agent-specific implementation"""
        raise NotImplementedError

    def record_action(self, event_type, **kwargs):
        """Standardized action recording"""
        self.database_manager.insert_action(
            event_type=event_type,
            **kwargs
        )

Individual Agent Details

BrowserUse
DoBrowser
MultiOn
Agent E
Skyvern

BrowserUse Agent

Type: AI-powered browser automation Provider: Open-source library Models: GPT-4, Claude, local LLMs

Key Features

Natural language understanding
Visual recognition capabilities
Automatic error recovery
Multi-step reasoning

Configuration

# collector/.env
OPENAI_API_KEY=sk-...
ANTHROPIC_API_KEY=sk-ant-...

# Optional configuration
BROWSERUSE_MODEL=gpt-4-vision-preview
BROWSERUSE_TEMPERATURE=0.7

Strengths

Excellent at understanding complex instructions
Handles dynamic content well
Good dark pattern detection
Robust error handling

Limitations

Requires API credits
Can be slower for simple tasks
May hallucinate interactions

Best Use Cases

Complex multi-step workflows
Sites with dynamic content
Testing natural language understanding
Dark pattern susceptibility testing

Agent Selection Guide

Choose BrowserUse when:

You need natural language understanding
Testing complex multi-step tasks
Evaluating AI agent capabilities
Cost-effective API usage

Choose DoBrowser when:

Testing requires real browser environment
Need cookie/session persistence
Testing with browser extensions
Simulating authenticated users

Choose MultiOn when:

Dealing with visual-heavy interfaces
Need cross-site automation
Testing e-commerce flows
Requiring robust form handling

Choose Agent E when:

Maximum dark pattern detection needed
Complex reasoning required
Need detailed execution traces
Enterprise-grade reliability

Choose Skyvern when:

Testing non-standard interfaces
Pure visual approach needed
Canvas or WebGL applications
Cross-platform testing

Performance Benchmarks

Agent	Avg Task Time	Success Rate	DP Detection	Resource Usage
BrowserUse	45s	85%	78%	Medium
DoBrowser	60s	75%	65%	High
MultiOn	40s	88%	80%	Low
Agent E	90s	92%	95%	Medium
Skyvern	120s	70%	82%	High

Adding Custom Agents

To add a new agent to LiteAgent:

Create agent class inheriting from WebAutomationBase:

# collector/web_automation_custom.py
from collector.web_automation_base import WebAutomationBase

class CustomWebAutomation(WebAutomationBase):
    async def run_task(self):
        # Implement agent logic
        pass

# collector/web_automation_factory.py
elif agent_method == "custom":
    return CustomWebAutomation(...)

Add Docker support:

# Dockerfile.custom
FROM python:3.11
# Agent-specific setup

Update configuration:

# docker-compose.yml
custom:
  build:
    dockerfile: Dockerfile.custom
  # Configuration

Agent Capabilities Matrix

Capability	BrowserUse	DoBrowser	MultiOn	Agent E	Skyvern
Click Elements	✅	✅	✅	✅	✅
Type Text	✅	✅	✅	✅	✅
Screenshots	✅	✅	✅	✅	✅
Scroll	✅	✅	✅	✅	✅
Hover	✅	✅	✅	✅	⚠️
Drag & Drop	⚠️	✅	✅	⚠️	❌
File Upload	✅	✅	⚠️	✅	❌
iFrames	✅	✅	✅	✅	⚠️
Popups	✅	✅	✅	✅	⚠️
Multi-tab	⚠️	✅	✅	⚠️	❌

Legend: ✅ Full Support | ⚠️ Partial Support | ❌ Not Supported

Next Steps

Dark Patterns

Learn about the dark patterns agents are tested against

Agent Setup

Detailed setup instructions for each agent

Running Tests

Start testing agents with LiteAgent

Getting Started

Core Concepts

Setup & Configuration

Running Tests

Output & Analysis

Agent Types

Supported Agents Comparison

Agent Architecture

Individual Agent Details

BrowserUse Agent

Key Features

Configuration

Strengths

Limitations

Best Use Cases

Agent Selection Guide

Choose BrowserUse when:

Choose DoBrowser when:

Choose MultiOn when:

Choose Agent E when:

Choose Skyvern when:

Performance Benchmarks

Adding Custom Agents

Agent Capabilities Matrix

Next Steps

Dark Patterns

Agent Setup

Running Tests

Getting Started

Core Concepts

Setup & Configuration

Running Tests

Output & Analysis

​Agent Types

​Supported Agents Comparison

​Agent Architecture

​Individual Agent Details

​BrowserUse Agent

​Key Features

​Configuration

​Strengths

​Limitations

​Best Use Cases

​Agent Selection Guide

​Choose BrowserUse when:

​Choose DoBrowser when:

​Choose MultiOn when:

​Choose Agent E when:

​Choose Skyvern when:

​Performance Benchmarks

​Adding Custom Agents

​Agent Capabilities Matrix

​Next Steps

Dark Patterns

Agent Setup

Running Tests

Agent Types

Supported Agents Comparison

Agent Architecture

Individual Agent Details

BrowserUse Agent

Key Features

Configuration

Strengths

Limitations

Best Use Cases

Agent Selection Guide

Choose BrowserUse when:

Choose DoBrowser when:

Choose MultiOn when:

Choose Agent E when:

Choose Skyvern when:

Performance Benchmarks

Adding Custom Agents

Agent Capabilities Matrix

Next Steps