pai-browser-use 0.1.0

Pydantic AI Toolsets for browser use

pai-browser-use

Pydantic AI Toolsets for browser automation using Chrome DevTools Protocol (CDP).

Inspired by browser-use, designed for Pydantic AI agents.

Features

• Browser Automation Tools: Navigation, state inspection, interaction, and element queries
• Multi-Modal Screenshots: Automatic image splitting for long pages with ToolReturn support
• Type-Safe CDP Integration: Direct access to cdp-use API with full type hints
• Fully Tested: Comprehensive test suite with Docker-based Chrome container

Installation

Use pip:

pip install pai-browser-use

Or use uv:

uv add pai-browser-use

Quick Start

Prerequisites

Start a Chrome instance with CDP enabled:

# Option 1: Using Chrome directly
google-chrome --remote-debugging-port=9222

# Option 2: Using Docker container
./dev/start-browser-container.sh

Basic Usage

import os
from pydantic_ai import Agent
from pai_browser_use import BrowserUseToolset

agent = Agent(
    model="anthropic:claude-sonnet-4-5",
    system_prompt="You are a helpful assistant.",
    toolsets=[
        BrowserUseToolset(cdp_url="http://localhost:9222/json/version"),
    ],
)

result = await agent.run("Find the number of stars of the wh1isper/pai-browser-use repo")
print(result.output)

See examples/agent.py for a complete example.

Available Tools

Navigation (4 tools)

• navigate_to_url, go_back, go_forward, reload_page

State Inspection (5 tools)

• get_page_info, get_page_content, take_screenshot, take_element_screenshot, get_viewport_info

Interaction (4 tools)

• click_element, type_text, execute_javascript, scroll_to

Query (3 tools)

• find_elements, get_element_text, get_element_attributes

Logging

The project includes detailed INFO level logging for debugging and development. By default, only ERROR logs are shown.

Enable Detailed Logging

# Set log level via environment variable
export PAI_BROWSER_USE_LOG_LEVEL=INFO

# Then run your script
python your_script.py

Available Log Levels

• ERROR (default): Only show errors
• WARNING: Show warnings and errors
• INFO: Show detailed operation flow
• DEBUG: Show all debugging information including actual data

What Gets Logged

INFO level shows:

• CDP connection establishment
• Browser target creation/reuse
• Tool execution lifecycle
• Page navigation steps
• Screenshot capture and processing
• Element interactions and queries
• Session state updates
• Resource cleanup

DEBUG level additionally shows:

• Extracted text content (first 500 characters)
• HTML content preview (first 500 characters)
• Element details (tag, text, attributes, position)
• JavaScript execution results
• Tool call arguments and return types
• Full page information (URL, title, viewport)
• All intermediate data for debugging

Examples

INFO Level - See operation flow:

import os
os.environ["PAI_BROWSER_USE_LOG_LEVEL"] = "INFO"

from pai_browser_use import BrowserUseToolset

async with BrowserUseToolset(cdp_url="http://localhost:9222/json/version") as toolset:
    # Shows what operations are being performed
    pass

DEBUG Level - See actual data:

import os
os.environ["PAI_BROWSER_USE_LOG_LEVEL"] = "DEBUG"

from pai_browser_use.tools.state import get_page_content

# Will show the actual text/HTML extracted
content = await get_page_content(content_format="text")
# DEBUG log shows: "Text content preview (first 500 chars): ..."

Development

# Install dependencies
uv sync

# Run tests
pytest tests/

# Run example
python examples/agent.py

# Try DEBUG logging demo (shows extracted content)
PAI_BROWSER_USE_LOG_LEVEL=DEBUG python demo_debug_logging.py

License

BSD 3-Clause License - see LICENSE for details.