use std::env;

use agcore::init_tracing;
use agcore::llm::{
    cycle::{CycleConfig, LlmCycle},
    provider::openai::OpenaiProvider,
    types::{ChatResponse, OpenaiContentPart},
};

fn extract_response_text(response: &ChatResponse) -> &str {
    match &response.message {
        agcore::llm::types::OpenaiChatMessage::Assistant { content, .. } => match content {
            agcore::llm::types::ContentField::String(s) => s,
            agcore::llm::types::ContentField::Array(parts) => {
                for part in parts {
                    if let OpenaiContentPart::Text { text } = part {
                        return text;
                    }
                }
                "[无文本内容]"
            }
        },
        _ => "[非 assistant 消息]",
    }
}

#[tokio::main]
async fn main() {
    dotenvy::dotenv().ok();
    init_tracing();

    let api_key = env::var("OPENAI_API_KEY").expect("未设置 OPENAI_API_KEY 环境变量");
    let base_url = env::var("OPENAI_BASE_URL").expect("未设置 OPENAI_BASE_URL 环境变量");
    let model = env::var("OPENAI_MODEL").expect("未设置 OPENAI_MODEL 环境变量选择所要使用的模型");

    let provider = OpenaiProvider::new(base_url, api_key, model.clone());
    let config = CycleConfig {
        model,
        max_tokens: Some(65536),
        temperature: Some(1.3),
        ..CycleConfig::default()
    };

    let mut cycle = LlmCycle::new(Box::new(provider), config)
        .with_system_prompt("你是一个简洁的助手，对于任何问题都是用一句话回答。".to_string());

    println!("发送请求...");

    match cycle.submit("介绍一下你自己吧。".to_string(), vec![]).await {
        Ok(response) => {
            println!("LLM 回复：{}", extract_response_text(&response));
            println!(
                "Token 用量：{} 输入， {} 输出",
                response.usage.prompt_tokens, response.usage.completion_tokens
            );
        }
        Err(e) => {
            eprintln!("请求失败：{e}");
            std::process::exit(1);
        }
    }
}