zclaw_openfang/desktop/src-tauri/src/classroom_commands/chat.rs

//! Classroom multi-agent chat commands
//!
//! - `classroom_chat` — send a message and receive multi-agent responses
//! - `classroom_chat_history` — retrieve chat history for a classroom

use std::sync::Arc;
use tokio::sync::Mutex;
use serde::{Deserialize, Serialize};
use tauri::State;

use zclaw_kernel::generation::{
    AgentProfile, AgentRole,
    ClassroomChatMessage, ClassroomChatState,
    ClassroomChatRequest,
    build_chat_prompt, parse_chat_responses,
};
use zclaw_runtime::CompletionRequest;

use super::ClassroomStore;
use crate::kernel_commands::KernelState;

// ---------------------------------------------------------------------------
// State
// ---------------------------------------------------------------------------

/// Chat state store: classroom_id → chat state
pub type ChatStore = Arc<Mutex<std::collections::HashMap<String, ClassroomChatState>>>;

pub fn create_chat_state() -> ChatStore {
    Arc::new(Mutex::new(std::collections::HashMap::new()))
}

// ---------------------------------------------------------------------------
// Request / Response
// ---------------------------------------------------------------------------

#[derive(Debug, Clone, Serialize, Deserialize)]
#[serde(rename_all = "camelCase")]
pub struct ClassroomChatCmdRequest {
    pub classroom_id: String,
    pub user_message: String,
    pub scene_context: Option<String>,
}

// ---------------------------------------------------------------------------
// Commands
// ---------------------------------------------------------------------------

/// Send a message in the classroom chat and get multi-agent responses.
// @connected
#[tauri::command]
pub async fn classroom_chat(
    store: State<'_, ClassroomStore>,
    chat_store: State<'_, ChatStore>,
    kernel_state: State<'_, KernelState>,
    persistence: State<'_, crate::classroom_commands::persist::ClassroomPersistence>,
    request: ClassroomChatCmdRequest,
) -> Result<Vec<ClassroomChatMessage>, String> {
    if request.user_message.trim().is_empty() {
        return Err("Message cannot be empty".to_string());
    }

    // Get classroom data
    let classroom = {
        let s = store.lock().await;
        s.get(&request.classroom_id)
            .cloned()
            .ok_or_else(|| format!("Classroom '{}' not found", request.classroom_id))?
    };

    // Create user message
    let user_msg = ClassroomChatMessage::user_message(&request.user_message);

    // Get chat history for context
    let history: Vec<ClassroomChatMessage> = {
        let cs = chat_store.lock().await;
        cs.get(&request.classroom_id)
            .map(|s| s.messages.clone())
            .unwrap_or_default()
    };

    // Try LLM-powered multi-agent responses, fallback to placeholder
    let agent_responses = match generate_llm_responses(&kernel_state, &classroom.agents, &request.user_message, request.scene_context.as_deref(), &history).await {
        Ok(responses) => responses,
        Err(e) => {
            tracing::warn!("LLM chat generation failed, using placeholders: {}", e);
            generate_placeholder_responses(
                &classroom.agents,
                &request.user_message,
                request.scene_context.as_deref(),
            )
        }
    };

    // Store in chat state
    {
        let mut cs = chat_store.lock().await;
        let state = cs.entry(request.classroom_id.clone())
            .or_insert_with(|| ClassroomChatState {
                messages: vec![],
                active: true,
            });

        state.messages.push(user_msg);
        state.messages.extend(agent_responses.clone());

        // Persist chat to SQLite
        if let Err(e) = persistence.save_chat(&request.classroom_id, &state.messages).await {
            tracing::warn!("[ClassroomChat] Failed to persist chat for {}: {}", request.classroom_id, e);
        }
    }

    Ok(agent_responses)
}

/// Retrieve chat history for a classroom
// @connected
#[tauri::command]
pub async fn classroom_chat_history(
    chat_store: State<'_, ChatStore>,
    classroom_id: String,
) -> Result<Vec<ClassroomChatMessage>, String> {
    let cs = chat_store.lock().await;
    Ok(cs.get(&classroom_id)
        .map(|s| s.messages.clone())
        .unwrap_or_default())
}

// ---------------------------------------------------------------------------
// Placeholder response generation
// ---------------------------------------------------------------------------

fn generate_placeholder_responses(
    agents: &[AgentProfile],
    user_message: &str,
    scene_context: Option<&str>,
) -> Vec<ClassroomChatMessage> {
    let mut responses = Vec::new();

    // Teacher always responds
    if let Some(teacher) = agents.iter().find(|a| a.role == AgentRole::Teacher) {
        let context_hint = scene_context
            .map(|ctx| format!("关于「{}」，", ctx))
            .unwrap_or_default();

        responses.push(ClassroomChatMessage::agent_message(
            teacher,
            &format!("{}这是一个很好的问题！让我来详细解释一下「{}」的核心概念...", context_hint, user_message),
        ));
    }

    // Assistant chimes in
    if let Some(assistant) = agents.iter().find(|a| a.role == AgentRole::Assistant) {
        responses.push(ClassroomChatMessage::agent_message(
            assistant,
            "我来补充一下要点 📌",
        ));
    }

    // One student responds
    if let Some(student) = agents.iter().find(|a| a.role == AgentRole::Student) {
        responses.push(ClassroomChatMessage::agent_message(
            student,
            &format!("谢谢老师！我大概理解了{}", user_message),
        ));
    }

    responses
}

// ---------------------------------------------------------------------------
// LLM-powered response generation
// ---------------------------------------------------------------------------

async fn generate_llm_responses(
    kernel_state: &State<'_, KernelState>,
    agents: &[AgentProfile],
    user_message: &str,
    scene_context: Option<&str>,
    history: &[ClassroomChatMessage],
) -> Result<Vec<ClassroomChatMessage>, String> {
    let driver = {
        let ks = kernel_state.lock().await;
        ks.as_ref()
            .map(|k| k.driver())
            .ok_or_else(|| "Kernel not initialized".to_string())?
    };

    if !driver.is_configured() {
        return Err("LLM driver not configured".to_string());
    }

    // Build the chat request for prompt generation (include history)
    let chat_request = ClassroomChatRequest {
        classroom_id: String::new(),
        user_message: user_message.to_string(),
        agents: agents.to_vec(),
        scene_context: scene_context.map(|s| s.to_string()),
        history: history.to_vec(),
    };

    let prompt = build_chat_prompt(&chat_request);

    let request = CompletionRequest {
        model: "default".to_string(),
        system: Some("你是一个课堂多智能体讨论的协调器。".to_string()),
        messages: vec![zclaw_types::Message::User {
            content: prompt,
        }],
        ..Default::default()
    };

    let response = driver.complete(request).await
        .map_err(|e| format!("LLM call failed: {}", e))?;

    // Extract text from response
    let text = response.content.iter()
        .filter_map(|block| match block {
            zclaw_runtime::ContentBlock::Text { text } => Some(text.as_str()),
            _ => None,
        })
        .collect::<Vec<_>>()
        .join("");

    let responses = parse_chat_responses(&text, agents);
    if responses.is_empty() {
        return Err("LLM returned no parseable agent responses".to_string());
    }

    Ok(responses)
}