Any page or document, reduced to what answers
CrawlGenie Read turns PDF to markdown for LLM pipelines, and does the same for any page or file. Hand it a link or a file and a question. It reads the whole thing, keeps the passages that answer, and tells you how many tokens it saved. Scanned PDFs are read with OCR; Google Docs and Sheets links just work; personal data can be hidden before the text reaches a model.
5–15% of the page
A question keeps only the matching passages; sentence mode keeps only the matching lines and table rows.
Every common file type
PDF, scans, DOCX, XLSX, PPTX, ODT, RTF, images, Google Docs, Sheets, Slides, YouTube captions.
Personal data stays home
Emails, phones, card numbers and IDs can be replaced before anything leaves the platform.
What happens when you use Read?
Give it a link or a file
Paste a URL, drop a file, or let ChatGPT or Claude pass it through the connector.
Ask a question (optional)
"What is the refund window?" "Revenue by segment, pages 30–60." Without a question you get clean text.
Get passages and a receipt
What answers, where it came from, and the tokens kept versus avoided.
What do people use Read for?
Contracts, policies, reports
Answer from the right pages of long documents instead of uploading them whole.
Research and due diligence
Read ten sources for one question under one budget.
Chat users
Through the connector, a PDF link costs a few hundred tokens instead of the whole file.
What you get, in detail
Most of what an agent is asked about is not a web page: it is a PDF, a scanned contract, a spreadsheet, a slide deck, a Google Doc someone shared, or a recorded call on YouTube. CrawlGenie Read parses all of these on the server without a model. PDFs are read from their text layer with lines and tables rebuilt by position; scans go through OCR automatically when the text layer is empty; Word, Excel, PowerPoint and OpenDocument files are parsed directly; Google Docs, Sheets and Slides links are converted to their export form; YouTube links return the caption track as timestamped paragraphs.
The question is what makes it cheap. With a question, Read keeps only the passages that answer, ranked without a model, and sentence mode keeps only the matching lines and table rows. Page ranges avoid parsing a 400-page filing to answer from the appendix. Personal data such as emails, phone numbers, card numbers and identifiers can be replaced before the text leaves the platform, with counts reported. The response always says how many tokens were kept and how many the full document would have cost, and the same call works from ChatGPT and Claude through the connector.
For developers
POST /v1/scrape {url, q} for pages, POST /v1/parse for files, POST /v1/answer {urls, q} for several at once. Options: pages, granularity, redactPII, cache. Web scraping for LLMs · PDF to markdown.
What would this save you?
Move the sliders. The saving is computed from the measured ratios above and the model price you pick.
Assumes 10% of the page is kept per read (measured range 5–15%). Receipts in the product show the real number for your traffic.
Questions people ask
Does it read scanned PDFs?
Yes. When a PDF has no text layer, OCR runs automatically inside the platform.
Does it transcribe video?
YouTube videos with captions, yes. Audio or video files without captions, no.
Spend tokens on answers, not on pages.
Create a workspace in a minute. Connect ChatGPT, Claude, Cursor or your own agents. Every answer comes with a token receipt.