Upload Files
Uploading files is the fastest way to give your agent structured knowledge. Manuals, policy documents, product catalogues, help articles, and data exports can all be uploaded and indexed in minutes.
Supported file formats
| Format | Extensions | Notes |
|---|---|---|
| Text-layer PDFs only. Scanned image PDFs (no text layer) will produce poor results — use a text-based PDF or export from the source application. | ||
| Word Document | .docx, .doc | Headings, paragraphs, and tables are all extracted. Embedded images are skipped. |
| Plain Text | .txt | Raw text files. Simplest format — no formatting overhead. |
| CSV | .csv | Tabular data. Each row is treated as a searchable unit. Best for product lists, FAQs in spreadsheet form, or data tables. |
| Excel Spreadsheet | .xlsx, .xls | All sheets are extracted and indexed. Cell values are read; charts and images are skipped. |
How file processing works
When you upload a file, it goes through a three-stage pipeline automatically:
1
Upload and extraction
The file is stored securely and its text content is extracted. For PDFs this reads the text layer; for DOCX this reads paragraphs and tables; for CSV/XLSX this reads cell values row by row.
2
Chunking and embedding
The extracted text is split into overlapping chunks (each chunk is a passage of several hundred words). Every chunk is embedded into a vector using the all-MiniLM-L6-v2 model and stored in the Qdrant vector database. The system records how many chunks and tokens were created per file.
3
Indexed and live
Once all chunks are embedded, the source status changes to Completed. From this point the file is active in retrieval — the agent will search it when answering questions.
Status indicators
Each file in your knowledge base shows a status badge:
- •Pending — file is queued, processing has not yet started.
- •Processing — chunks are being created and embedded. Large files may take a minute or two.
- •Completed — file is fully indexed and live. You can see how many chunks and tokens were generated.
- •Failed — processing encountered an error. The most common cause is a scanned PDF with no text layer, a password-protected file, or a corrupted document.
Uploading files
1
Open your agent and go to Knowledge Base
In the left menu, click Knowledge Base then select the Files tab.
2
Upload your files
Click the upload button or drag and drop files directly. You can upload multiple files at once.
3
Monitor status
Each file shows its status badge. Completed files are immediately active in retrieval.
Updating and deleting files
To update a file, delete the existing version and re-upload the new one. When a file is deleted, all its vector embeddings are immediately removed from the index — the agent stops drawing on that content at once.
Tips for the best results
- •Use text-based PDFs, not scanned image PDFs. Export from the original application (Word, Google Docs) rather than scanning a printout.
- •Add clear headings and section titles. The chunking algorithm respects headings, so well-structured documents produce more accurate retrieval.
- •Keep files focused on a single topic where possible. A 10-page product FAQ retrieves better than a 200-page company handbook where every topic is mixed together.
- •For very large documents (100+ pages), consider splitting them into topic-specific files — the agent will retrieve only the relevant chunk regardless, but smaller files make status tracking easier.
- •CSV files work well for structured data like pricing tables, product attributes, or FAQ lists — one question/answer per row.
💡
After uploading a file, test the agent immediately with a question that should be answered by that file. If the answer is wrong or missing, check that the status is Completed and that the file contains searchable text.
Was this page helpful?