### Answer Here are some strategies to optimize token usage in conversation flow agents: 1. Keep prompts concise Avoid verbose prompts. One practitioner noted watching out for token usage when formatting prompts—sometimes pasting entire large prompts increases token consumption unnecessarily. [(1) ]() 2. Avoid excessive agent "yapping" If your agent talks too much, it increases tokens. You can add instructions like "no yapping" in your response guidelines to keep responses shorter. 3. Use LLM States / Conversation Flow nodes effectively Breaking conversations into distinct nodes with focused prompts helps reduce context drift and keeps each interaction's token footprint smaller. [(2) ]() 4. Knowledge base considerations Knowledge base retrieval may only incur token usage when actually called. You can configure

kb_config

parameters like

top_k

and

filter_score

to control how much data is retrieved:

json
"kb_config": {
  "top_k": 3,
  "filter_score": 0.6
}

, [(3) ]() 5. Format knowledge base data efficiently Using markdown formatting and splitting knowledge base content into smaller, well-organized chunks (rather than large tables or PDFs) can help the LLM retrieve information more efficiently. [(4) ]() 6. Adjust model temperature Lower

model_temperature

values (closer to 0) can produce more deterministic, potentially shorter responses. [(2) ]() , [(3) ]()

​ To learn more, see the sources I considered: [[1] Part 1: BuildYour First Retell Voice Al Agent That Sells Cars]() [[2] Create Conversation Flow](https://docs.retellai.com/api-references/create-conversation-flow) [[3] Get Conversation Flow](https://docs.retellai.com/api-references/get-conversation-flow) [[4] Master The Future of Voice AI - Conversation Design & Agent Persona]() ​

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