Tool call integration

[caffeinism]

I'm able to do tool calls via langchain. But can this library support the case of injecting tool schema from the client? I want the client to receive input about the tool call, not the executed result. The limitation says it doesn't support it, but I'm wondering if it's technically impossible, or if it's possible but complicated. When I tried logging .bind_tools, it passes the schema itself to the langchain just fine, while the function call in the response is intercepted.

[samuelint]

You are referring to this limitation?

Functions cannot be passed through open ai API. Every functions need to be defined as a tool in langchain. Usage Example

I think it is feasible but not implemented, would require to have langgraph to select the tool from the openai API schema.