Skip to main content

Manage and delete files

Open File management via the menu. Here you manage uploads and files the AI has created. Selecting a file to manage is something different from adding it as a source to a chat; see Chat with documents.

Uploading files​

Drag files into the Drop files here box or click Choose files. While uploading, you will see the progress there; you can add more files with the upload button in the toolbar.

Deleting multiple files at once​

  1. Find the files using the filter at the top and, if needed, choose files or images.
  2. Use the table view for an overview with checkboxes. You can also tick files in the grid view.
  3. Tick the files you want to delete. In the table, the checkbox above the column selects the rows on the current page; it does not automatically select all pages.
  4. Check the number of selected files. A selection may include files you previously ticked. Use Clear selection in the table if you want to start over.
  5. Click the trash can at the top. It appears as soon as you have selected something.
  6. Check the number in the confirmation and confirm only if the selection is correct.
  7. Wait for the notification and the refreshed overview. In case of a partial error, the notification will state how many files were and were not deleted. Check the remaining files before trying again.

Deleting removes the selected files from File management. Keep your own copy first if you may need it later. Turning off an attachment in chat does not delete the saved file.

Downloading and saving​

Select one file to download it directly. With Download as ZIP you can download a selection together. If you have configured a Drive connection, you can also send supported files to the linked folder. You set up these connections under Preferences → Drive connections.

Checking processing​

An upload is only usable once processing is complete. Documents may be up to 150 MB per file; separate limits apply to audio, video, and images. An accepted upload does not yet mean that every format or every type of content can be processed.

If there is an error, first read the error message. A supported document can be processed again via the action menu, or by selecting one document and choosing Reprocess. If necessary, resubmit a damaged or protected document as a readable file. Reprocessing replaces the previous text extraction.

For failed audio and the option to remove the source recording after transcription, see Audio and transcription. In that case, the transcription text can be retained; this is different from deleting the entire file in File management.

Checking which sources were read​

Below an answer, Sources can show which content the model used: Full text, Selection, Summary, Partially processed, or Not read. An attached file is therefore not automatically read in full. Check this information before using an answer that must cover all parts of a file.

With a selection or shortened result, you can ask more specifically for a chapter or the remaining parts. If processing is incomplete, first check the file status. An assistant administrator can choose Full source text required under Source text: in that case, summaries are not used as a substitute for the text. Large sources may be read in parts and use more tokens. Even then, checking the sources message remains necessary.

With Automatic, the assistant receives small files (up to 5 files totaling about 30,000 words) in full with every question. That gives faster answers. In larger files, the assistant looks up the relevant sections.

Protected PDFs and missing text​

A protected PDF that opens without a password and contains a text layer can be processed. The text layer is read; OCR of protected scanned pages is not supported. A PDF that requires a password or contains only scanned pages must be resubmitted as a readable, unprotected file.

Check the processing status after uploading. A file that has not yet been processed or does not yield usable text is not silently treated as read. Below the answer, Sources can indicate that processing is incomplete, with the reason. Only ask the AI for conclusions about the full content after processing has succeeded.

On this page