Upload anything confusing and get it explained in plain English.
PDFs and images are sent to a server to be read. Word, Excel and PowerPoint files are opened here and only their text is sent. Nothing is stored.
A PDF, a photo, a screenshot, a Word, Excel or PowerPoint file, or plain text. Up to 8MB. Nothing to choose first — it works out what it is.
Reading it…
PDFs and images are sent to a server to be read. Word, Excel and PowerPoint files are opened in this browser tab and only their text is sent. What is sent is used to answer that one request and is not stored. Nothing is kept after the answer comes back, there is no copy to come back to later, and no document here is ever used to train a model.
Some documents are written to be understood and some are not. A hospital bill lists codes instead of treatments, a lease buries the part about leaving in clause 14, and an error message names a file you have never opened. Drop any of them here and this says what it is, who wrote it for whom, and which parts you would regret not reading.
There is nothing to choose first. It works out what the file is — a contract, a bill, a spreadsheet, a screenshot, a stack trace — and explains it the way that kind of document needs explaining. Then you can ask it questions about what it just told you.
It reports how sure it is and what it could not read, which matters more than it sounds. A scanned page with no text layer, a table cut off at the margin and a language it cannot follow all look like nothing at all from the inside, and a summary that skips them silently is worse than one that says so.
It explains; it does not advise. It will tell you that a clause renews the agreement automatically unless you cancel sixty days ahead, and what to ask a solicitor about it. It will not tell you whether to sign.
Medical bills, utility statements and insurance paperwork are written for the system that produced them. This pulls out what was charged, what somebody else already paid, and what the number at the bottom is actually asking you for.
A lease, a job offer, a supplier agreement. The parts worth finding are always the same — how it ends, how it renews, what happens if you are late — and they are never in the same place twice.
A dashboard, an error dialog, a settings page, a chart from a deck. Drop the picture in and it reads what is on screen and says what it is showing.
It explains in the language the document is written in, so run it, then ask your follow-up questions in whichever language you prefer to read the answer in.
A parent's letter from a pension provider, a friend's tenancy agreement. The suggested questions are the useful part here: they are what somebody who reads these for a living would ask next.
A PDF, a photo or screenshot, a Word, Excel or PowerPoint file, or plain text and code. Up to 8MB. Nothing to configure and nothing to describe first.
What the document is, why it exists, who it was written for, and how long it would take to read. Then the parts that matter, the dates, the terms worth knowing, and anything you need to do.
Every explanation carries how confident it is and why, plus a list of what it could not read. Read that before you rely on the rest of it.
The suggested questions are a starting point — tap one to put it in the box, or write your own. Answers come from your document, and it says so when the answer is not in there.
Copy the explanation or download it as Markdown. Nothing is kept here, so if you want it later, this is the moment.
PDFs and images are, yes — they are sent to be read and that is said above the dropzone rather than buried here. Word, Excel and PowerPoint files are different: those are opened in this browser tab and only the text inside them is sent, because the model reads text, PDFs and pictures and nothing else. Either way, nothing is stored once the answer comes back.
Yes, which is why every explanation says how confident it is and what it could not read. The failure worth watching for is not a wrong fact — it is a confident summary of a document whose middle pages were a scan it could not see. That is what the unknowns list is for. Treat it as a first read by somebody quick, not as a second opinion.
No, and it is built not to. It explains what a document says and what to ask somebody qualified about it. Anything that amounts to legal, medical or financial advice is a question for a person who can be held to their answer.
Usually, yes — a photograph of a page or a scanned PDF is read the way a person reads it rather than by pulling out a text layer. Handwriting is more variable. Where it cannot make something out it says so instead of guessing, and if you only need the text itself, Image to Text is the more direct tool.
Six per document, and the counter beside the box shows what is left for today across all the AI tools here. There is no account and no payment behind any of it, so the limit is what keeps it free and working for everybody rather than a nudge toward buying something.
The pages for contracts, medical bills and code errors pass a hint about what you were looking for, but the hint never overrides the file. If you drop a payslip on the contract page it will tell you it is a payslip. That is deliberate — a tool that explains your document as the thing you searched for is agreeing with you rather than reading.
Not here, not yet. One file at a time. For text you already have in front of you, Text Compare shows exactly what changed between two versions.