NavoChatAI workspace
Maintain and test AI knowledge
Import, review, update, and test sources so agents and Copilot use reliable information.
Updated 2026-08-04For: AI managers · Content managers
Knowledge does not determine whether business content is correct. The team remains responsible for source quality, structure, and updates.
Steps
Create a knowledge base
Separate content by business boundary, such as product policy, support process, and technical guidance, instead of putting everything in one collection.
Add sources
Upload supported files or write FAQs, maintain aliases, and use clear names. During upload, choose a preset such as automatic, document, or semantic chunking.
Review and apply changes
Confirm successful status and inspect summaries or failures. After adding, replacing, or deleting a document, wait for processing and choose Apply changes to sync the searchable version.
Test retrieval
Enter a realistic question in the knowledge base detail view and run the recall test. Review matched chunks, scores, and whether each result entered context. When changes are pending, the test uses the last applied version.
Grant access
Connect the knowledge base to the required agent or Copilot configuration and test retrieval with realistic phrasing.
Advanced search settings
Knowledge search settings control semantic caching, initial candidate count, chunk size, chunk overlap, and citation thresholds. Global settings apply to all knowledge bases; an individual knowledge base can override its recall thresholds. Run the recall test again after changing these settings.
Test questions retrieve the correct source, answers match it, and outdated content is removed.