Documents
Upload files to a knowledge base and follow their indexing state.
Opening a knowledge base shows its documents. From here you upload new files, check how their indexing is going and act on the ones that failed.
Uploading documents
You can select several files at once or drag them onto the upload area. The accepted formats and maximum size are defined by the platform, and the file picker already comes configured with them, so you do not have to remember them.
Every file is validated before it is sent. Files that fail validation are flagged and not sent, but they do not block the rest: the valid ones carry on.
While files upload you will see a progress list separate from the table. Files that finish successfully disappear from that list after a few seconds, because they are already in the table below. Files that fail stay until you dismiss them: they are the only trace of a file that never got created.
Indexing consumes tokens from your account with the provider, because it uses the knowledge base's credential. Long files, or files with many pages, consume more.
Document states
Uploading is asynchronous: the file is accepted straight away and processed in the background. You can keep working meanwhile — the table updates on its own.
| State | What it means | What to do |
|---|---|---|
| Pending | Accepted and queued, not processed yet | Wait |
| Processing | Text is being extracted and indexed | Wait |
| Ready | Indexed and available for queries | Nothing |
| Failed | The process did not finish | Check the reason and retry |
When a document fails
A failed document shows the reason next to its state: what happened and what you can do about it. The most common causes are:
- A PDF with no text layer. It is a scanned PDF — an image of a document. The platform does not do optical character recognition, so there is no text to extract. The fix is to upload a version with selectable text.
- Problems with the provider credential, usually an expired key or an account with no credit.
Retrying indexing
The retry action asks you to select the file again. That is not an oversight: if the base does not keep originals, the platform does not store the file after indexing it, so it cannot fetch it back on its own.
Retrying keeps the document, deletes its previous passages and starts over. That is why you use this rather than uploading the file again, which would create a duplicate document.
It also works to reindex a document that is already ready, for example after you have updated its contents.
You cannot retry while a document is pending or processing: you have to wait for it to finish.
Downloading the original
If the knowledge base keeps originals, stored documents get a download action in their row.
The download appears document by document, not for the whole base: a single table can hold documents uploaded before and after the option was turned on, and only the latter have a stored file.
The download link is temporary and generated the moment you click, so there is no point in copying it or saving it for later.
Deleting a document
Deleting a document also removes its indexed passages: it stops appearing in searches immediately.
Next steps
Once your documents are Ready you can query them: start with Search.
Updated on Aug 04, 2026