Hello,
This is my first post on this forum, but I’m not new DLAI and I have lost count of the courses I’ve taken. Thank you!
The A2A Protocol short course is clearly a winner in my book! The instructors are wonderful and I’ve never had such a clearer understanding of A2A in a “short course”.
I’ve worked on all the exercises by myself writing agents that instead of invoking LLMs, invoke fully functional Langflow and Flowise workflows that are powered by SLMs on the ollama backend. I’m happy to report that each of these agents are accessible via their individual A2a server containers. 
I see SLMs have their limits and therefore, I have been trying to build the Orchestrator agent using AWS Bedrock and this is where I’m hitting a roadblock.
I’m able to use both, the ChatModel and the AwsBedrockChatModel to instantiate and run inference. However, when I use the llm object within a RequirementAgent I run into this error “unsupported model or your request does not allow prompt caching”.
Clearly, I’ve confirmed the models are accessible and supported. And I also tried to implement UnlimitedCache and SlidingCache as ChatModelOptions, but that did not make things better. From Bedrock documentation it seems that prompt caching (aptly named) needs to be part of the prompt.
Question(s)
- How do I package prompt caching instructions in the prompt of a RequirementAgent?
- And if not,what is the fix for this issue?
- I looked for AWS Bedrock examples for BeeAI framework, but did not find any beyond what I’ve already done. Am I missing something?
Thank you in advance!
Andy