### Answer Based on the available sources, I don't have explicit documentation about how Retell AI caches or reuses tool call results across turns. What I can share from the documentation: - The

transcript_with_tool_calls

field captures when (at what utterance, which word) a tool was invoked and what the result was [(1) ]() [(2) ]() - Function calling allows the LLM to "intelligently choose to output a JSON object containing arguments to call one or many functions" [(3) ]() The behavior you're describing—where the agent proposes new time slots without re-calling the calendar tool—is likely driven by the underlying LLM's decision-making. The LLM decides when to trigger a function call based on the conversation context . If the LLM determines it can respond using information already in the conversation history (including previous tool results) , it may not issue a new tool call. To ensure the agent always fetches fresh data, you could: - Adjust your prompt to explicitly instruct the agent to re-check availability before offering new slots - Design your function description to clarify when the tool should be called Check out these potentially useful sources: - [Integrate Function Calling]() - [LLM WebSocket]() Would you like me to provide a general answer about how LLMs typically handle tool result caching?

​ To learn more, see the sources I considered: [[1] LLM WebSocket](https://docs.retellai.com/api-references/llm-websocket) [[2] Update Call](https://docs.retellai.com/api-references/update-call) [[3] Integrate Function Calling](https://docs.retellai.com/integrate-llm/integrate-function-calling) ​

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

how LLMs typically handle tool result caching? @AI Evy

### Answer Based on the available sources, I don't have specific documentation about how LLMs handle tool result caching internally. However, I can share some general context: LLMs maintain **context memory**—they are "able to maintain coherent conversations across multiple turns by referencing earlier dialogue." [(1) ]() This means the model can reference information from earlier in the conversation, including results from previous tool calls that appear in the conversation history. The key point is that LLMs don't have a dedicated "cache" for tool results. Instead: 1. **Conversation context**: Previous tool call results become part of the conversation transcript, which the LLM uses as context for generating responses 2. **Decision-making**: The LLM decides whether to call a tool again based on its interpretation of the conversation and prompt instructions—not a formal caching mechanism This is why adjusting your prompt to explicitly require fresh data retrieval (as mentioned in my previous response) is the recommended approach when you need the agent to always re-call tools rather than rely on earlier results in the conversation history. Would you like me to point you to additional resources on function calling behavior?

​ To learn more, see the sources I considered: [[1] Choosing the Best LLM for Your AI Voice Agents | Retell AI](https://www.retellai.com/blog/choosing-the-best-llm-for-your-voice-ai-agents) ​

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