Adding Documents

This article covers uploading files to a knowledge base that already exists, and following them until they are indexed. For what a knowledge base is, see Knowledge Base. To create one, see Creating a Knowledge Base. To crawl web pages, see Adding a Website.

Two of the four document types, Advanced Documents and Data Table, need a Document Processing LLM on the knowledge base. Without one, their cards are greyed out and hovering either asks for Advanced Document Processing settings. You set this in the knowledge base's Document Processing section, described in Creating a Knowledge Base.

The examples use ApplianceBrands, a knowledge base holding support content for two appliance brands, Vantera and Solmar. It defines two tags, brand and product.

Choosing a document type

Open the knowledge base. It opens on its Data Source section. Click + New Data Source. The Add New Data Source screen shows five cards, with Websites selected by default. The other four are document types.

Uploading documents

Text-Only Documents, Advanced Documents and Data Table all use the same upload screen.

1. Drop box: Drag files onto it, or click it to choose them. One upload takes up to 20 files.

2. Supported formats: The formats this type accepts, plus the size and count limits that apply to your tenant.

To add the two brand manuals to ApplianceBrands:

1. Select Text-Only Documents.
2. Add vantera-refrigerator-support.txt and solmar-refrigerator-support.txt.
3. Each file has its own + Add tag control. Tag the first file brand: vantera and product: refrigerator, and the second brand: solmar and product: refrigerator. For how tags narrow retrieval in a flow, see Knowledge Base Tags.
4. Click Save. Both files appear in the Documents table and are processed in the background.

Replacing a document

A knowledge base does not pick up changes to a file on its own. To update a document, upload the new version under the same file name (upper and lower case count as the same). Once you confirm, it replaces the existing document and is processed again.

You cannot replace a document while it is Processing or Pending, and you cannot put two files with the same name in one upload.

Uploading FAQ Data

FAQ Data has no + Add tag control. Its tags come from the spreadsheet itself.

1. Banner: Gives the column rule, which is explained below.

2. Click to download the template.: Downloads a workbook whose headings already match this knowledge base's tags.

3. Drop box: Accepts .xlsx files.

Only the first sheet of the workbook is read. Its columns are read like this:

Here is a file for ApplianceBrands:

⚠️ Warning: Question and Answer must be the first two columns. The upload screen accepts a file that has them anywhere, but processing then fails and the file shows Failed.

Following processing

1. + New Data Source: Opens the Add New Data Source screen.

2. Websites: The websites added to this knowledge base. See Adding a Website.

3. Documents: Every uploaded document, with its Tags, Document Type, Size and Modified Date. A document tagged on upload shows each tag with its value, such as brand: vantera. An FAQ workbook shows tag names only, because its values differ from row to row.

4. Train Status: How far processing has got. The statuses are explained in the table below.

5. Download and delete: The download icon returns the original file. The trash icon deletes the document and its indexed content, after a confirmation.

While any source is Pending or Processing, the knowledge base's settings sections are read-only. Tags stay editable.

The knowledge base itself has a status too, shown by the Status icon on the Knowledge Bases list:

⚠️ Warning: A single Failed source keeps the whole knowledge base Not ready. While it is Not ready, a flow's Knowledge Retrieval or RAG Agent node cannot select it, and you cannot publish the project, because publishing requires every knowledge base to be Ready. Fix or delete the failed source.

Refresh Index

Refresh Index in the Data Source section runs processing again, but only for sources that Failed or that show the settings-changed info icon. Indexed sources are left alone, and an Indexed website is not re-crawled. To refresh a website, see Adding a Website.

• The button appears when at least one source is Failed or shows the info icon, and no source is Pending.
• It is disabled while any source is Processing.
• It starts right away, with no confirmation, and the knowledge base stays Not ready until processing finishes.

Limits

These are the defaults. The upload screen shows the size and count limits that apply to your tenant. Websites do not count toward any of them.