Building Knowledge Bases and Recall Testing

Start with Getting Started with Knowledge Bases for the complete create → import → test → use workflow. This page points advanced workflows to the relevant steps.

Creating and Importing Materials

Choose the step you need below; begin with creation if this is your first knowledge base.

1. Open [Knowledge Base] → [New Knowledge Base]

See Create a Knowledge Base. An embedding model is optional: you can start with keyword search and add embeddings when needed.

2. Select Data Sources

See Add and Organize Material for files, folders, notes and URLs. For scanned documents, see Document Preprocessing.

3. Wait for Processing to Complete

Check the text and Chunks for missing or garbled content. See Check Material and Recall for the acceptance steps.

4. Perform Recall Testing

Use questions you already know the answers to, then inspect the source and complete context. Follow Recall Testing before binding the knowledge base to an Agent.

Validate Recall Results with Real Questions

A finished import is not enough: the correct source and the conditions needed to answer must appear in the results. The recall guide explains how to diagnose missing, truncated or outdated results.

Understanding RAG Settings

For embedding models, reranking, Top K and chunking, use Model and Retrieval Settings. Change one setting at a time and repeat the same questions; re-index existing material when changing chunking settings.

Use Case: Building an Internal Policy Q&A Library

Keep current policies together, test real questions, and check sources before using the library in an Agent. See Knowledge Base Cases and Using with Agent for the full workflow. Update or remove outdated policies and repeat recall testing when the source material changes.

Why are there no results after importing?

Check processing status and extracted text first; if an embedding model is configured, also check its connection. Follow Knowledge Base Troubleshooting to locate the failing step.