Custom Tool Output and Use It in Agent

[projectssimm]

Hi guys,

Here is the situation, I have 3 tools to use one by one. But, the outputs of which are not in the same types. Say:

Tool 1: output normally, and can be further thought by GPT in the chain,
Tool 2: output specially, which means the output should not be thought further by GPT again, because in that way the output format and data will be not correctly,
Tool 3: output normally, like Tool 1.

In this way, I found it hard to get the correct answer for myself. If all the tools are set to return_direct=False, the answer to Tool 2 will not be right, and if I set return_dircect=True, the chain of Tools 1->2->3 would be lost...

Also, I want to use the tools in an agent, such as:

tools = [Tool1, Tool2, Tool3]

agent = initialize_agent(tools, llm, agent=AgentType.CONVERSATIONAL_REACT_DESCRIPTION, verbose=True, memory=memory, return_intermediate_steps=True)

So for these tools, especially tool2, what should I do?

Any help will be highly appreciated.
Best