Where to put import statements when defining a tool?

[lemorage]

I checked out many examples of tool usage in Letta framework, and it seemed that, when defining tools, they all had import statements inside the function body. Is it just a good practice or convention to do so, or is it a must? If it is a must, then could you explain the reasons behind it?

[HG2407]

+1, suppose I want to make a field optional, I need to add the line "from typing import Optional" outside the function, I can do that through ui, but how to do that with python

[meir23]

What happens is that letta takes all the code related to the tool and does the following:

1. It converts all the code to a string and runs it in its own dedicated environment when the time comes. Therefore, if you run the tool from your code base, you often put the import command inside the method. So that it will definitely become a string and when reading and running this function in the letta environment, it will be with the relevant import.

2. Inside the method, you make a Docstring to guide the language model on how to convert it into a schema. Remember that it is of utmost importance here to describe:

• What the method does and when it is worth using it. I would also add a few shots.
• If there are parameters to the function, then describe each parameter and what its type is
• What the result should be

And finally: In ADE, you have the option of putting all the code that is relevant to a particular tool and then you simply paste the code. This way, the import can also be outside. What happens is that the system automatically knows how to connect it to your client. From there, it's a very short way to connect it to your specific agent.

[HG2407]

ok, but my problem is I don't have access to the ade due to running it on cloud. so I need to find a way to add the import statements while creating a tool using client.create_tool, is there a way ? Or I should actually just provide the stringified version of the function directly by modifying the code ?