Agentics2.0 Full Release

[gliozzo]

Includes Substantial Changes

1. Transducible Functions
2. Explainability
3. Dynamic Composition
4. Cleaned up tutorials
5. Removed Applications

[junkyul]

In PydanticTransducerCrewAI, adding cache=False and rpm=200 may reduce transduction failure due to LLM server

• rpm limites the number of calls per miniute.
• I observed a bursty behavior, while transducing 90 questions in bird, it was slient for a long time and finished all transduction in 15 seconds. This causes trouble when I was trying to transduce 1500 questions at once.

agent = Agent(
            role=self.prompt_params["role"],
            goal=self.prompt_params["goal"],
            backstory=self.prompt_params["backstory"],
            verbose=verbose,
            max_iter=max_iter,
            llm=self.llm,
            tools=tools if tools else [],
            cache=False
        )
        self.crew = Crew(
            agents=[agent],
            tasks=[task],
            process=Process.sequential,
            verbose=verbose,
            manager_llm=self.llm,
            function_calling_llm=self.llm,
            chat_llm=self.llm,
            cache=False,
            rmp=200
        )

[junkyul]

In Documentation on Transducible Function,
importance concepts need definitions.

• ordinary functions
• structural transparency

In addition, this is a question.

• slot-level provenance: the definition implies it is 1:1 mapping between slots in the input and the output type. Is it necessary? In the end, one could apply semantic merge like operations on the fields to yield a single field but this needs clarification since it could also be 1:N or N:M, which we might want to exclude.