Data Management
Pages, files and datasets in a single repository, in folders you arrange. Drag anything to re-file it. Deleting is reversible for 30 days, with an Undo the moment you do it.
The toggle is per item, and it only makes something selectable. Nothing reaches the model until you actually pick it in the composer. Meters at the top show storage used, how many items are on the menu, and roughly how many characters that adds up to — so context is a number you can see rather than a surprise.
Pages written in the Hub, plus .txt, .md, .csv, .tsv, .json, .html, .xml, .yaml and .log — up to 400,000 characters each.
PDF, Word and Excel are queued for the reader and turned into text, including scanned pages, usually within a minute or two. It shows “Reading this document” until it is done.
Photos, archives and video upload and download normally, and are marked as not available to the AI — because nothing here can turn them into text. They still play and preview.
Two limits we would rather you heard from us. A document over about 4.5 MB is refused with a message saying so, and PowerPoint is not read yet. Reading a document costs tokens and appears in your usage as content_extract, like any other AI work.
Wrong type, too large, or no readable text — the item tells you which, and gives you Try again. It does not sit spinning while you assume it is working. The dot beside a file means it is hidden from the AI, and the tree explains that at the bottom rather than leaving you to guess.
Upload a .csv or .tsv and it becomes a dataset: previewed as a table in the Hub, downloadable in full, and usable as the source of a Data Visualization. 16 MB per file, 20 GB across the account — and the error names which one you hit.
Tell us what you are working on and we will open an account.
Request access