AI Assistant

The AI Assistant lives in your dashboard under AI Agent. It does three jobs from one chat box:

1. Answers product questions - "How do I scrape a JavaScript page?", "What does the Walmart Product API return and what does it cost?" - grounded in this documentation and the live product catalog.
2. Gets the data for you - "get me the price and rating of this Amazon product", "find coffee shops in Berlin on Google Maps", "what does this page say". It runs the matching product API itself, answers with the values you asked for, says what the run cost and offers the full result as a file to download right in the chat.
3. Prepares bigger runs - a batch of URLs or a crawl of a site. It fills in the parameters and shows a card with the estimated cost, complete curl, Node.js and Python code, and a button that opens the form filled in; you review it and start it yourself.

The assistant is available to the Owner, Admin, Developer and Manager roles of a workspace, and it can only run what your role may run.

A question in the chat runs at most two charged calls, each a single request to one target - the same call the REST API would make, at the same price. It never starts a batch or a crawl on its own: those turn one sentence into hundreds of charged pages, so they always wait for your click. Nothing a scraped page says can make the assistant run anything: only your own message can.

How answers work

When you ask a question, the assistant retrieves the most relevant passages from these docs and answers from them, citing the pages it used. This is retrieval-augmented generation: the reply is built from the current documentation, not from memory, so it stays in sync with the product.

If the docs do not cover something, the assistant says so rather than inventing an answer. The chat itself never costs credits.

What it can do

The assistant works through a small set of tools. Each reply may use one or more of them:

Tool	What it does	Credits
Search docs	Finds and quotes the relevant documentation passages.	Free
Product details	Reads one product from the live catalog: what it returns, its price, endpoint, every parameter with allowed values, and a working example request.	Free
Get data	Runs one product API for you - the same tools the MCP server offers: scrape_url, google_search, google_maps_search, amazon_product, booking_search, yelp_business, walmart_product and others.	The product's own price
Prepare request	Fills in a request of any Scraper API in the catalog, a Web Scraping API request, a batch or a crawl as a card you start yourself.	Free until you start it
Get file	Shows the beginning of a file your workspace already produced - paste its id from Files.	Free

Under a reply that fetched data the chat shows what the run did - how many requests it made, what it charged and how long it took - with a button per format of the result (JSON, CSV, Markdown and so on) and a link to the run's record in Files, where every format and its retention live.

The assistant checks every parameter before it shows a card - the parameter names and values the product form accepts, and for a batch or crawl the same rules the API applies - so the form opens with valid values. While it works, the chat shows Assistant is typing… and the step it is on (looking up a product, preparing the request); if an answer fails or times out, Try again asks the same question once more.

From card to result

1. The card shows what will run and the estimated cost, plus complete cURL, Node.js and Python code for the same request. Copy it and replace YOUR_SCRAPEIT_API_KEY with your key.
2. Open the card: the product's Playground, the batch page or the crawler page opens with the request filled in.
3. Check or change anything, then start it there. It runs exactly like a request you set up by hand: the same SSRF protection, concurrency limits, shared cache and credit prices.

Only roles that can run that kind of request get a card for it - for example, a role without Web Scraping API access gets no playground card. For no-code scrapers the assistant explains the scraper and links its page in the dashboard.

Credits and limits

• Questions are free. Documentation search, product lookups, file lookups and prepared cards never charge credits.
• Fetched data costs what the API costs. A reply that ran a product API charges that product's price, exactly as the REST route does, and says so on the card. A call that is blocked or fails on our side is not charged, as everywhere else (billing rules). One question runs at most two charged calls.
• You start batches and crawls. A request you start from a card costs what the same request costs from the API or the dashboard; an estimate is not a charge.
• Daily conversation budget. To keep the assistant responsive for everyone in the workspace, chat has a per-organization daily token budget. When it is used up, the chat replies with a limit notice and resets the next day; your scraping credits and API access are unaffected. Very long answers or large files use more of it.
• Daily message limit. Each workspace may send a set number of chat messages per day (UTC), 100 by default. Past that the chat answers with AI_CHAT_DAILY_REQUEST_LIMIT_EXCEEDED and the message is not stored; the count starts again at 00:00 UTC. Scraping credits and the API are unaffected. Need more? Contact support.

Example prompts

The chat sidebar ships ready-made prompts you can click to try each capability:

Get data in the chat - the assistant runs the product API and answers with the data, what it cost and a file to download:

• Read https://books.toscrape.com/ and list the books on its first page
• Get the price and rating of the Amazon product B09B8V1LZ3
• Find coffee shops in Berlin on Google Maps and list the five best rated

Fill in a form for me - you get a card; its Open button opens the Playground, Batch or Crawler page with every field filled in, and you press Run there:

• Prepare a batch of https://books.toscrape.com/ and https://quotes.toscrape.com/
• Set up a crawl of books.toscrape.com up to depth 2
• Fill in an Airbnb search for Paris in December
• Fill in a Yellow Pages search form for pizza places in New York
• Open Amazon Best Sellers for Electronics in the Playground

Help with the API - ready-to-run code and how products, parameters and integrations work:

• Show me a Python example for Google SERP
• Give me code that extracts the title and price from https://books.toscrape.com/catalogue/a-light-in-the-attic_1000/index.html
• What does the Walmart Product API return and how much does it cost?
• How do I connect ScrapeIt to Claude or Cursor with MCP?
• How do I set up a webhook and verify its signature?

Advice:

• Which proxy type should I use for a protected e-commerce site?
• When am I charged credits, and what counts as a billable result?
• My requests to a site come back TARGET_BLOCKED - what should I change?

Privacy and safety

• Your conversations are yours. Other members of the workspace cannot see or continue them.
• History is kept for 6 months. A conversation is deleted automatically 6 months after its last message.
• Secrets stay masked. API keys, tokens, signing secrets and passwords in URLs are masked before anything is sent to the language model and before your message is stored; storage locations are never shown.
• Page content is treated as data. Text from your files reaches the model as untrusted content: instructions written inside a page are not followed, and they can never start a request.

Related pages

• Billing & credits - exactly what a run costs.
• Web Scraping API - the request a card prepares.
• Website Crawler API - the crawl jobs it can prepare.
• Files & artifacts - where results land.