Skip to main content
This function uses AI and incurs costs.
This function has multiple overloads
Extracts structured data from web pages using AI-powered content analysis.This function provides intelligent data extraction from web pages using various strategies including HTML parsing, image analysis, and Markdown conversion. Or by using Text or Image Content. It supports extraction from entire pages or specific elements, with built-in caching and retry mechanisms.
async def extract_structured_data(
    *,
    source: Page | Locator,
    data_schema: type[BaseModel] | dict[str, Any],
    prompt: str | None,
    strategy: Literal['IMAGE', 'MARKDOWN', 'HTML'],
    model: str,
    api_key: str | None,
    enable_dom_matching: bool | None,
    enable_cache: bool | None,
    max_retries: int | None,
) -> Any
Extract data from web pages or specific elements using HTML, IMAGE, or MARKDOWN strategies with DOM matching support.

​
Features and limitations

Features:
  • Smart Caching: Hashes inputs and uses KV Cache for persistent storage
  • DOM Matching: With enable_dom_matching=True, values match DOM elements for smart caching
  • Multiple Strategies: HTML, IMAGE, or MARKDOWN based on content type
  • Flexible Models: Use any up-to-date model from anthropic, openai or google based on your needs.
Limitations:
  • Model Variability: Quality varies by model - experiment to find the best fit
  • DOM Complexity: Dynamic structures can affect caching and matching
  • IMAGE Constraints: Cannot capture truncated or off-screen content
  • Schema Design: Complex schemas may reduce accuracy

​
Examples

from typing import TypedDict
from playwright.async_api import Page
from intuned_browser.ai import extract_structured_data
class Params(TypedDict):
    pass
async def automation(page: Page, params: Params, **_kwargs):
    await page.goto('https://books.toscrape.com/catalogue/a-light-in-the-attic_1000/index.html')
    # This will extract the book details from the page, using the HTML strategy with the gpt-4o model.
    # The data_schema is a JSON Schema dictionary that defines the structure of the data to extract.
    # You can also use a Pydantic BaseModel instead of a JSON Schema dictionary.
    book = await extract_structured_data(
        source=page,
        strategy="HTML",  # The HTML strategy is the default strategy and will be used if no strategy is provided.
        model="gpt-4o",
        data_schema={
            "type": "object",
            "properties": {
                "name": {"type": "string"},
                "price": {"type": "string"},
                "description": {"type": "string"},
                "in_stock": {"type": "string"}, # Must be string because we enabled DOM matching
                "rating": {"type": "string"}
            },
            "required": ["name", "price"]
        },
        prompt="Extract book details from this page",
        enable_cache=True,  # since this is True, the method will call AI for the first time, and then whenever you call this method it will return cached results as long as the DOM is the same.
        enable_dom_matching=True,  # since this is True, the method will return the results mapped to the DOM elements, you MUST enable cache for this to work.
        max_retries=3
    )
    print(f"Found book: {book['name']} - {book['price']}")

​
Arguments

​
source
Page | Locator
required
Playwright Page object to extract data from the entire page or Locator object to extract data from a specific element.
​
data_schema
type[BaseModel] | dict[str, Any]
required
Schema defining the structure of the data to extract. Can be either a Pydantic BaseModel class or a JSON Schema dictionary.
​
prompt
str
Optional prompt to guide the extraction process and provide more context. Defaults to None.
​
strategy
Literal['IMAGE', 'MARKDOWN', 'HTML']
Type of extraction strategy:
  • “HTML” (default) - Best for text-heavy pages with structured content
  • “IMAGE” - Best for visual content, charts, or complex layouts
  • “MARKDOWN” - Best for article-style content with semantic structure
​
enable_dom_matching
bool
Whether to enable DOM element matching during extraction. When enabled, all types in the schema must be strings to match with the DOM elements. Extraction results are mapped to their corresponding DOM elements and returned with matched results. These results are intelligently cached, allowing subsequent extractions with minor DOM changes to utilize the cached data for improved performance. Defaults to False.
​
enable_cache
bool
Whether to enable caching of extraction results. Defaults to True.
​
max_retries
int
Maximum number of retry attempts on failures. Failures can be validation errors, API errors, output errors, etc. Defaults to 3.
​
model
str
AI model to use for extraction. Defaults to “claude-haiku-4-5-20251001”.
​
api_key
str
Optional API key for AI extraction (if provided, will not be billed to your account). Defaults to None.

​
Returns: Any

The extracted structured data conforming to the provided schema.