要点
The usual formats
PDF, Word, Excel, PowerPoint, OpenDocument and RTF are read directly – no converting, no intermediate step.
Ask instead of skim
“What does it say about the notice period?” beats skimming 40 pages. The answer comes with the passage it relies on.
Compare several documents
Put three quotes side by side and have them turned into a table – including a note when one compares apples to oranges.
Honest about scans
If a PDF is only a photo without a text layer, Qwirbel says so – instead of inventing something that sounds plausible.
What can be read – and what cannot
Above is which files work. Here is where the limit sits and what happens to the document.
Which formats
PDF, Word, Excel, PowerPoint, OpenDocument and RTF. You drop the file into the chat and Qwirbel works with it – summarise, answer questions about it, pick out passages, put content into another form.
⚠ A pure scan has no text
A photographed or scanned contract is at first only an image to the computer – there is no text layer in it that could be read. Qwirbel says so openly instead of guessing. Anyone regularly handling such documents needs text recognition first.
What happens to the document
It stays on your machine. It is not uploaded, not staged at a provider and not used for training. That is exactly the point with contracts, medical records or personnel files – papers you may not or will not give to a cloud service.
How much at once
That depends on the model's context window, and that in turn on the card. A very long document can exceed the window – Qwirbel then works section by section. At 90 percent it saves first instead of quietly truncating.
When it should go into knowledge
Documents you need often do not belong in every chat but in your knowledge store. The agent then searches them itself when needed, without you attaching them each time.
关于这一点大家会问
Does the document leave my machine?
No. Reading happens locally. Only if you deliberately chose an external model for the answer does that excerpt go there – the file itself stays with you.
Can it read scanned paper?
Only if a text layer exists. For a pure scan it says so honestly instead of guessing. Such files need text recognition first.
How large may the files be?
There is no plan limit. The practical limit is how much text the chosen model can hold at once – for very long documents it pays to ask about specific sections.
May I use this with client data?
That is exactly what local operation is for: there is no processor involved because the data never leaves the building. Whether a specific case is permissible remains your decision – we don't give legal advice.
这个 Qwirbel 也能做
所有本领都住在同一个程序里——买一次，一把钥匙。