Skip to content

Upload files

Supported file types, the plan-based size limit, the safety checks every upload passes, and what FILE_TOO_LARGE and UNSUPPORTED_FILE_TYPE mean.

Open a chatbot's Knowledge tab and choose Upload file. One file becomes one knowledge source, titled with the file's name.

Supported types

Exactly five: PDF, DOCX, TXT, MD (also .markdown), and CSV.

Anything else is rejected before it is stored:

Common cases that hit this: .doc (the pre-2007 Word format — save as .docx), .xlsx (export the sheet as CSV), .pptx, and images. There is no OCR, so a PDF that is a scan of paper contains no extractable text; it will upload and produce nothing useful.

Size limit

The maximum upload size is set by your plan, not by the product, and the error names your own limit:

The number in that message is your plan's, so read it rather than assuming. Oversized uploads are rejected from the declared request size before the body is read, so a very large file fails fast instead of uploading and then failing.

A second, different FILE_TOO_LARGE can come from a PDF that is small on disk but unsafe to expand:

That is a bound on extraction, not on the byte count. Split the document and upload the parts.

What happens to an upload

  1. 1
    The extension is checked against the five supported kinds.
  2. 2
    The size is checked against your plan.
  3. 3
    The bytes are checked against the extension — a .pdf that is not really

    a PDF is rejected even though its name looked right.

  4. 4
    The file is scanned for malware.
  5. 5
    Only then is it stored and queued for processing.
Note:

Because the content check reads the actual bytes, renaming a .doc to .docx does not get it through. Re-save it in the real format instead.

After the upload, the source appears as processing until the worker has extracted, chunked, and embedded it. See Knowledge base overview.

Updating a document

Files are not versioned. To publish a corrected document, delete the old source and upload the new one — otherwise both versions stay retrievable and the agent may quote either.

Where to go next

Last updated