Upload documents
Upload files to a knowledge base, follow their indexing states, and handle tables and scanned pages.
Upload the files your agent should answer from. Baseera reads them, splits them into passages, indexes them and keeps the original private.
Supported files
PDF, DOCX, Markdown, TXT, HTML and CSV, up to 50 MB each. Uploading a file with the same name replaces it with a new version.
Upload files
Open Knowledge, then open the knowledge base. Press Add source in the bar above the page, or use the Upload files area.
Drop files on Drop files here, or click to choose, or click to pick them.
Optionally open Who can read the files you add and write access rules. Leave it empty for everyone. See Knowledge access.
Watch each file move through Waiting, Uploading, Finishing and Uploaded. Then follow it in the Sources table.
The file goes straight to private storage and never passes through a Baseera web server. You can leave the page. It checks progress every few seconds while a source is in progress.
Read the indexing states
Each row in Sources has a Status:
| Status | Meaning |
|---|---|
| Uploading | The file is being sent. |
| Queued | Waiting for a worker. |
| Parsing | The text is being read out of the file. |
| Embedding | Passages are being turned into vectors with the knowledge base's key. |
| Ready | Searchable. |
| Failed | Something stopped it. The row shows why and a Retry button. |
The knowledge base itself shows Empty, Processing, Needs attention or Ready. Sources that are still processing are not searched.
Replace a file without a gap
When you upload a file under a name that already exists, the previous version keeps answering, and its link keeps working, until the new version is ready. A failed upload loses nothing. While it processes, the row shows Updating…, and "The previous version (vN) is still in use."
Tables and scanned pages
PDF and DOCX files go through a document parser that understands layout, including tables and Arabic text. CSV, Markdown, TXT and HTML are read directly.
Scanned PDFs are read twice. Baseera first reads the text layer. If a page has almost no text, it reads the pages again with OCR in Arabic and English. A source read this way is marked OCR with the note "Text read from scanned pages; check quotes against the original." Review OCR sources before you rely on exact quotes.
If a PDF has no readable text even with OCR, it fails with "No text found. If this is a scanned PDF, it has no readable text." Upload a text version instead. The agent's Overview lists such sources under Needs attention with the hint "No text could be read from it, even with OCR. Upload a text version."
Use Test retrieval to confirm that a number from a table comes back in the passage. See Knowledge.
What can go wrong
| Message | Cause and fix |
|---|---|
| "Unsupported file type." | Use PDF, DOCX, Markdown, TXT, HTML or CSV. |
| "Larger than 50 MB." | Split the file. |
| "The file is empty." | Check the file before uploading. |
| "Storage limit reached." | The organization has a storage limit for original files. Delete sources you no longer need. |
| "A newer version was uploaded." | Another upload replaced this one while it was in progress. |
| "PDF and DOCX files can't be read yet: the document parser isn't set up." | The parser is not available in this environment. Ask your Baseera administrator. |
| "The file couldn't be read. It may be damaged or protected." | Remove the password or re-export the file. |
| "Too large: it would make more than 5,000 chunks. Split it into smaller files." | Split the document. |
| "The provider rejected the key. Replace the key, then retry." | Fix the embedding key in the knowledge base Embedding settings, then press Retry. |
| "The key is out of credit. Add credit with the provider, then retry." | Top up, then Retry. |
| "The provider is rate limiting this key. Try again in a few minutes." | Wait, then Retry. |
| "Processing didn't finish. Try again." | Press Retry. |
| "The embedding model returned vectors of an unexpected size, so nothing was stored." | Choose another model or size in the knowledge base settings. |
Rate limits and provider errors are retried automatically up to three times before a source shows Failed.