BrowserTool

Use BrowserTool class to create a new tool that interacts with web browsers. This tool can be used to automate web browsing tasks, such as opening a webpage, clicking on elements, and extracting information from web pages.

from npiai import BrowserTool
 
class MyTool(BrowserTool):
    def __init__(self):
        super().__init__(
            name='<Provide the name of your tool here>',
            description='<Provide a brief description of your tool here>',
            system_prompt='<Provide a system prompt for LLMs to interact with your tool here>',
        )

Attributes

name

• Description: The name of the tool.
• Returns: str

description

• Description: A brief description of the tool.
• Returns: str

system_prompt

• Description: A system prompt for LLMs to interact with the tool.
• Returns: str

tools

• Description: A list containing the JSON schema of the functions registered with the tool.
• Returns: List[ChatCompletionToolParam]. See the Schema Generation section for more information.

Sync Methods

add_tool

• Usage:

  tool.add_tool(tool_a, tool_b, agent_a, agent_b)

• Description: Adds tools or agents to the tool.

• Arguments:

  • *tools (BaseTool): The tools to add. They can be either a Tool or an Agent returns by the npiai.agent.wrap function.

• Returns: None

use_hitl

• Usage:

  tool.use_hitl(hitl_handler)

• Description: Attach the given HITL(human-in-the-loop) handler to this tool and all its sub-tools.

• Arguments:

  • hitl (HITLHandler): The HITL handler to attach.

• Returns: None

Async Methods

start

• Usage:

  await tool.start()

• Description: Starts the tool.

• Arguments: None

• Returns: None

end

• Usage:

  await tool.end()

• Description: Ends the tool.

• Arguments: None

• Returns: None

call

• Usage:

  results = await tool.call(llm_response.tool_calls)

• Description: Calls the function corresponding to the LLM tool calls and returns the results.

• Arguments:

  • tool_calls (List[ChatCompletionMessageToolCall]): The tool calls from the LLM response.

• Returns: List[ChatCompletionMessageToolCall]

  • role: "tool"
  • tool_call_id (str): The ID of the tool call.
  • content (str): The result of the corresponding function call.

Browser-related Methods

The BrowserTool class provides several methods to interact with web browsers. Here are some of the key methods:

get_text

• Usage:

  text = await tool.get_text()

• Description: Extracts the text content of the current web page as markdown.

• Arguments: None

• Returns: str

goto_blank

• Usage:

  await tool.goto_blank()

• Description: Go to about:blank page. This can be used as a cleanup function when a session finishes.

• Arguments: None

• Returns: None

get_screenshot

• Usage:

  screenshot = await tool.get_screenshot()

• Description: Takes a screenshot of the current web page.

• Arguments: None

• Returns: str | None: The base64-encoded image data or None if the screenshot fails.

get_page_url

• Usage:

  url = await tool.get_page_url()

• Description: Gets the URL of the current web page.

• Arguments: None

• Returns: str

get_page_title

• Usage:

  title = await tool.get_page_title()

• Description: Gets the title of the current web page.

• Arguments: None

• Returns: str

is_scrollable

• Usage:

  scrollable = await tool.is_scrollable()

• Description: Checks if the current web page is scrollable.

• Arguments: None

• Returns: bool

click

• Usage:

  await tool.click(element_handle)

• Description: Clicks on the specified element on the web page.

• Arguments:

  • element_handle (playwright.ElementHandle): The handle of the element to click. See the Playwright documentation (opens in a new tab) for more details.

• Returns: None

fill

• Usage:

  await tool.fill(element_handle, text)

• Description: Fills the specified text into the input field of the element on the web page.

• Arguments:

  • element_handle (playwright.ElementHandle): The handle of the input element. See the Playwright documentation (opens in a new tab) for more details.
  • text (str): The text to fill into the input field.

• Returns: None

select

• Usage:

  await tool.select(element_handle, value)

• Description: Selects the specified value from the dropdown list element on the web page.

• Arguments:

  • element_handle (playwright.ElementHandle): The handle of the dropdown list element. See the Playwright documentation (opens in a new tab) for more details.
  • value (str): The value to select from the dropdown list.

• Returns: None

enter

• Usage:

  await tool.enter(element_handle)

• Description: Presses the Enter key on the specified element on the web page.

• Arguments:

  • element_handle (playwright.ElementHandle): The handle of the element to press the Enter key. See the Playwright documentation (opens in a new tab) for more details.

• Returns: None

scroll

• Usage:

  await tool.scroll()

• Description: Scrolls the web page down by one viewport height to reveal more content.

• Arguments: None

• Returns: None

back_to_top

• Usage:

  await tool.back_to_top()

• Description: Scrolls the web page back to the top.

• Arguments: None

• Returns: None