A modern programming language that makes developers happier.

kotlinlang

image.png

Greetings, I'm trying to rewrite <https://academy.langchain.com/courses/deep-research-with-langgraph|this course python code> into `koog`. It is based on "agents", which are basically "nodes" in our case with shared context(thanks JetBrains &lt;3). I'm stuck writing one node that accepts `researchBrief` and should call *one* tool(tavily-search). I've used `requestLLMForceOnetool`, however the result is just a query to tool call, and I'm expecting the result of web search itself. Am I missing something? Should I `forwardTo` `nodeLLMRequest`?

<https://github.com/rcd27/deep-research-with-koog/blob/ed789f9db0b07b6a65b4fa733598b7ea34c67cea/app/src/main/kotlin/agent/strategy/deepResearchStrategy.kt#L45|link to code>

Hi Stanislav,
there's no need to forward the node to `nodeLLMRequest` , as `requestLLMForceOneTool` is a request itself.

You need to <https://docs.koog.ai/sessions/#request-methods-with-tools|process the received response>, like this:
```llm.writeSession {
            updatePrompt {
                system(researchQuestion.researchBrief)
            }
            val response = requestLLMForceOneTool(tavilySearchTool)
        }
    if (response is Message.Tool.Call) {
        // Handle tool call, e.g. with nodeExecuteTool() and then nodeLLMSendToolResult()
    } else {
        // Handle text response
    }
}```
Hope this answers your question!

Total success, thank you :slightly_smiling_face: I missed the concept of filtering `Message.Respone`  for it to be either `Tool.Call` or not

Viva open-source and community! :slightly_smiling_face: