Documents
Upload handbooks, SOPs, vendor contracts, and training material so your team can find them, and so the AI assistant can answer questions from them.
Page: Admin -> Documents (/admin/documents) · Role required: admin
Before uploads work, a platform admin must configure storage and an embedding provider once. See Document storage and embeddings.
Upload a document
- Go to Admin -> Documents.
- Click New.
- Enter a title and description.
- Upload the file.
- Save. The document appears in the table with its type and size.
| Supported for upload | Supported for AI indexing | Max size |
|---|---|---|
| pdf, docx, doc, xlsx, xls, csv, txt, png, jpg, jpeg, gif, webp | pdf, docx, doc, txt, csv | 25 MB |
Index a document for AI search
- Click the Index icon on the document row.
- UnitOps extracts the text, splits it into chunks of roughly 500 words with 50 words of overlap, and generates an embedding per chunk.
- When it finishes, the response reports the number of chunks and the model used, and the document is marked as indexed.
Re-index after replacing the file so the chunks match the new content.
Search your documents
- Expand the AI Document Search panel on the Documents page.
- Type a natural-language question, for example "How do I reset the fryer timer?".
- Choose the scope: all documents or one specific document.
- Choose how many results to return.
- Review the matching chunks and their similarity scores.
| Score | Meaning |
|---|---|
| 0.75 and above | Highly relevant |
| 0.50 to 0.75 | Moderately relevant |
| Below 0.50 | Weak match, consider rewording |
Remove an index
Click Remove Index. The embeddings are deleted and the document stops appearing in AI answers. The file itself stays in storage.
Use documents in chat
Once at least one document is indexed and an embedding provider is configured, the chat assistant automatically retrieves the most relevant chunks for a question and includes them in its context. There is no extra step for the user.
Privacy
Documents and their embeddings are scoped to your organization and enforced at the database level with row-level security. A user in another tenant can never retrieve your chunks, in the UI, in chat, or through the API.