Match a list of domain glob patterns against a URL. Args: domain_patterns: A list of domain patterns that can include glob patterns (* wildca

Match a page filter against a page

Some models like deepseek-reasoner dont allow multiple human messages in a row. This function merges them into one.

Derive the most reliable selector from recorder payload. Priority -------- 1. #id 2. Stable attrs: data-testid/data-test/data-au

Validate the output of the last action is what the user wanted

Ensures page is fully loaded before continuing. Waits for either network to be idle or minimum WAIT_TIME, whichever is longer. Also checks if t

Wrap text to fit within a given width. Args: text: Text to wrap font: Font to use for text max_width: Maximum width in pixels Ret

Execute an action

Add tool message to history

Clean up server resources.

Clean up all servers properly.

Convert input messages to a format that is compatible with the planner model

Extract JSON from model output, handling both plain JSON and code-block-wrapped JSON.

Final result from history

* Searches up to maxDepth levels upward and downward from a target node * for an interactive element. If an interactive element is found, returns it;

Get next action from LLM based on current state

Get a response from the LLM. Args: messages: A list of message dictionaries. Returns: The LLM's response as

Get the system prompt for the agent. Returns: SystemMessage: Formatted system prompt

Check if this is the last step

Check if the agent completed successfully - the agent decides in the last step if it was successful or not. None if not done yet.

Last action in history

List available tools from the server. Returns: A list of available tools. Raises: RuntimeError: If the serve

Read the given JSON file and return it as a Python dict.

Log the completion of the task

Get a description of the action for the prompt

Removes all highlight overlays and labels created by the highlightElement function. Handles cases where the page might be closed or inaccessible.

Remove last state message from history

Save conversation history to file.

Save current cookies to file

Save history to JSON file with proper serialization

Execute one step of the task

Returns a base64 encoded screenshot of the current page.

* Recursively builds a trimmed HTML string from the given node. * Only goes N layers deep. Deeper nested content is replaced with a placeholder. *

Get all unique URLs from history

Wait for the page to load completely. Args: timeout (int): Maximum time to wait for page load in milliseconds (default: 30000ms). Raise

Representation of the AgentHistoryList object

Add step number and goal overlay to an image.

Create a LangChain ChatOpenAI client with sane defaults. Parameters ---------- model : str OpenAI model name. Defaults to *gpt-4

Cancel current tasks that should be interruptible.

Remove markdown fences and whitespace from a reply string.

Convert dictionary-based actions to ActionModel instances

Convert input messages to the correct format

Convert messages for non-function-calling models

Converts simple XPath expressions to CSS selectors.

Creates a new browser context with anti-detection measures and loads cookies if available.

Creates a Pydantic model from function signature

Create initial frame showing the task.

Should not be thread blocking because posthog magically handles it

Execute a single step from history with element validation

Filter out sensitive data from the message

Generate a unique filename by appending (1), (2), etc., if a file already exists.

Initialize the message history with system message, context, task, and other initial messages

Initialize loop state attributes used for signal handling.

Input text into an element with proper error handling and state management. Handles different types of input fields and ensures proper element st

Input text into an element with proper error handling and state management. Handles different types of input fields and ensures proper element st

Load the prompt template from the markdown file.

Return a list of (path, text) for every .md file in *dir_path*.

Log the agent run

Replaces the sensitive data in the params

Run the planner to analyze state and suggest next steps

Get the version and source of the browser-use package (git or pip in a nutshell)

Setup dynamic action models from controller's registry