Upload documents the agent should learn from. PDF, Office, OpenDocument, EPUB, text, Markdown, JSON, and CSV.
Files are the right choice for content that does not live on a public web page: product manuals, policy documents, price sheets, internal FAQs you have exported, and anything a crawl cannot reach.
Open Knowledge, select Add, then Files.
What you can upload
| Extensions | |
|---|---|
.pdf | |
| Word | .doc, .docx |
| PowerPoint | .ppt, .pptx |
| Excel | .xlsx |
| OpenDocument | .odt, .ods, .odp |
| Rich text | .rtf |
| EPUB | .epub |
| Plain text | .txt, .csv |
| Markdown | .md, .markdown |
| JSON | .json |
Each file can be up to 10 MB. Upload several at once.
Word documents keep their heading structure when the file carries real heading styles, which gives the agent better context about where an answer came from. A document formatted with manually-enlarged bold text instead of Heading 1 reads as flat paragraphs.
How files are used
We read the text out of each file, break it into chunks, and store them the same way we store website pages. The agent searches those chunks at answer time and pulls in what is relevant. A file counts toward your knowledge chunk limit like any other trained source. See Limits.
Text beats scans
A PDF that is a scanned image with no selectable text has nothing for us to read, and the upload will tell you so. If you can select and copy the text in your PDF viewer, we can read it. If you cannot, run it through OCR first, export a text version, or paste the content as text.
Updating a file
To replace a file, upload the new version with the same source. It overwrites the old chunks rather than adding a second copy, so re-uploading a corrected document does not leave stale answers behind.