Agent support

[heepengpeng]

Expected Behavior
Implement Agent like in Langchain with different method， such as CoT，ReACT

[markpollack]

If we do something like this it would be a post 1.0 feature. FWIW, I found there to be varying definitions/implementations of 'agents', in particular I was able to reproduce langhchain/llamaindex demos of ReACT agent that did basic math (addition/subtraction/multiplcation) with basic functional calling behavior.

[kkolago]

I totally agree that implementing Agents would be very beneficial and would simplify writing more complex applications. I know that currently it is possible to model reasoning logic and interactions with external tools using function calling, advisors and evaluators, but the way you can model it in LangGraph by simply defining graph of though process is much cleaner. Also maybe creating some interfaces for implementing custom agents would be nice I think (to make it more opinionated).

It would be nice to have something similar in Spring AI afer 1.0 GA (especially that we already have some tools that possibly might be leveraged for reasoning graph building like Spring Statemachine).

[lambochen]

If we do something like this it would be a post 1.0 feature. FWIW, I found there to be varying definitions/implementations of 'agents', in particular I was able to reproduce langhchain/llamaindex demos of ReACT agent that did basic math (addition/subtraction/multiplcation) with basic functional calling behavior.

Have any relevant design documents or discussions?

[seethefruture]

Agree！ need agent support.

[shaohzhangebay]

need spring-ai-graph !