Skip to main content

​
CodeAgent Parameters

​
Core Settings

  • task: Task description string that defines what the agent should accomplish (required)
  • llm: LLM instance for code generation (required: ChatBrowserUse). If not provided, defaults to ChatBrowserUse()
  • browser: Browser session object for automation (optional, will be created if not provided)
  • tools: Registry of tools the agent can call (optional, creates default if not provided)
  • max_steps (default: 100): Maximum number of execution steps before termination
  • max_failures (default: 8): Maximum consecutive errors before termination
  • max_validations (default: 0): Maximum number of times to run the validator agent

​
Vision & Processing

  • use_vision (default: True): Whether to include screenshots in LLM messages. True always includes screenshots, False never includes screenshots
  • page_extraction_llm: Separate LLM model for page content extraction. You can choose a small & fast model because it only needs to extract text from the page (default: same as llm)

​
File & Data Management

  • file_system: File system instance for file operations (optional, creates default if not provided)
  • available_file_paths: List of file paths the agent can access
  • sensitive_data: Dictionary of sensitive data to handle carefully

​
Advanced Options

  • calculate_cost (default: False): Calculate and track API costs

​
Backwards Compatibility

  • controller: Alias for tools for backwards compatibility
  • browser_session: Alias for browser for backwards compatibility (deprecated, use browser)

​
Return Value

The run() method returns a NotebookSession object that contains:
  • cells: List of CodeCell objects representing each executed code cell
  • id: Unique session identifier
  • current_execution_count: Current execution count number
  • namespace: Dictionary containing the current namespace state with all variables

​
CodeCell Properties

Each cell in session.cells has:
  • id: Unique cell identifier
  • cell_type: Type of cell (‘code’ or ‘markdown’)
  • source: The code that was executed
  • output: The output from code execution (if any)
  • execution_count: Execution order number
  • status: Execution status (‘pending’, ‘running’, ‘success’, or ‘error’)
  • error: Error message if execution failed
  • browser_state: Browser state after execution

​
Example

session = await agent.run()

# Access executed cells
for cell in session.cells:
    print(f"Cell {cell.execution_count}: {cell.source}")
    if cell.error:
        print(f"Error: {cell.error}")
    elif cell.output:
        print(f"Output: {cell.output}")

# Access variables from the namespace
variables = session.namespace
print(f"Variables: {list(variables.keys())}")