Tool Calling

Learn the basic interface of tool calling within Orpheus.

For more advanced use cases, you will probably want to give the LLM a way to interact with its environment. Tool calling enables this by specifying a set of actions the model can take.

Orpheus supports tool calling via the Tool object builder, which you can then pass to the tools parameter of your request.

The response will then, if the model decides to use a tool, hold a vector of ToolCall objects that we can then use within our code.

empty_tool.rs

use orpheus::prelude::*;

fn main() {
    let client = Orpheus::from_env().unwrap();

    // Define a tool
    // For this example, it's just an empty tool we can use as a "switch".
    let my_tool = Tool::function("my_tool").empty();

    let res = client
        .chat("Call my tool")
        .model("openai/gpt-4o")
        .tools([my_tool]) // Pass an iterable of tools to the request
        .send()
        .unwrap();

    // `tool_calls` is a convenience method to
    // access tool calls in the response, if any
    if let Some(tool_calls) = res.tool_calls().unwrap() {
        for tool_call in tool_calls {
            println!("Tool call: {:?}", tool_call);
        }
    }
}

Tool call: Function { id: "call_dTmBXTw3BXhIbCL5r2dVvnnU", function: Function { name: "my_tool", arguments: "{}" } }