### Answer Based on the available documentation, when a knowledge base is linked to an agent, the agent will always try to retrieve information from the knowledge base before responding [(1) ]() . This is automatic behavior and happens for every response generation . Current behavior you're experiencing: - KB retrieval triggers on every turn regardless of whether the question needs KB content - This adds latency even when tools (like check availability) already have the answer Options to reduce unnecessary retrieval: 1. Adjust retrieval settings - You can configure how many chunks are retrieved via "Adjust KB Retrieval Chunks and Similarity" : - **Chunks to retrieve**: Set to 1-3 (default is 3) to reduce token overhead - **Similarity Threshold**: Increase from 0.6 to a higher value (like 0.7-0.8) so only highly relevant chunks match 2. Use Node-level KB (Conversation Flow only) - If you're using Conversation Flow agents, you can configure knowledge base at the node level . This means: - Only specific nodes that need KB access will retrieve from it - Nodes handling tool-based questions (like availability checks) can have no KB attached 3. Prompt engineering - Add instructions to your prompt telling the agent to prioritize tools over KB for specific question types. While this won't prevent retrieval, it may help the LLM process faster. Note: The documentation states retrieval latency is "generally under 100ms" , so if you're seeing 3-4 seconds, the delay may be coming from LLM processing of the retrieved chunks rather than retrieval itself. For Conversation Flow agents, node-level KB configuration is your best option to prevent retrieval on turns where it's not needed .

​ To learn more, see the sources I considered: [[1] Knowledge Base](https://docs.retellai.com/build/knowledge-base) ​

Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.