The community members are discussing how to dynamically orchestrate Bedrock agents, which require executing Lambdas to provide results. One community member suggests creating a class/function to create a new Lambda through Docker and a Lambda adapter, but notes the issue of plugging in dependencies. Another community member asks for an example of what is meant by "dynamically" and the original poster clarifies that they mean converting a small code fragment to be deployed to a Lambda at runtime, with some OpenAPI specifications being created. The discussion also touches on the similarities and differences between this approach and OpenAPI function calling, and the idea of having a large language model (LLM) create the function/Lambda at runtime. A community member has also created a related issue on GitHub.
I wonder how bedrock agents orchestration could be done dynamically. It does require lambdas to be executed to provide the results. Am open to suggestions.
the logic is kinda similar to openapi function calling, but difference here is bedrock will automatically perform the necessary calling for lambdas (think of them as the tools)
interesting, yeah was playing around with this recently out of necessity and it seems this is the closest thing the Bedrock enabled models have to any type of function calling... is the idea here to have the LLM create the function/lambda at runtime? as it's currently setup they basically do act as tools and seem to use some type of function calling AWS baked in somewhere as far as I can tell
Nice, reason I have stepped away for now is actually because the Knowledge Base feature (their baked in retrieval pipeline) is so janky, for another time..