OpenAI Web Automation
Pricing
$60.00 / 1,000 results
Controls a real browser with an OpenAI model to interact with web pages and extract structured data — no CSS selectors or page-specific scraping code required.
One or more web pages the browser will open before the AI agent begins interacting. Add multiple URLs to run the same task on several pages in sequence.
Instructions for what the AI agent should do on the page — e.g. search for a product, fill a form, or click through a checkout flow. The browser is already open; do not include navigation steps like 'go to URL'.
Describe the data you want extracted from the final page state. The AI will read the screenshot and page HTML, then return this data as JSON. Example: 'Extract the product title, price, and stock status.'
Optional. A JavaScript function that receives the Zod library (z) and returns a Zod schema. When provided, the extracted JSON will be validated and shaped to match the schema. Leave as (z) => null to use free-form JSON output. See https://zod.dev/ for schema syntax.
Maximum number of browser actions (clicks, typing, scrolling, etc.) the AI agent may perform before stopping. Increase for complex multi-step tasks; decrease to limit cost.
