const fs = require('fs');
const path = require('path');
const OpenAI = require('openai');

// read optional config file
const configPath = path.join(__dirname, 'llm_config.json');
const llmOptions = fs.existsSync(configPath) 
  ? JSON.parse(fs.readFileSync(configPath, 'utf8')) 
  : {};

console.log(`Read config from ${configPath}`);

function generateBotPersonality() {
    const personalities = fs.readFileSync(path.join(__dirname, 'bot_personalities.txt'), 'utf8')
      .split('\n')
      .map(personality => personality.trim())
      .filter(personality => personality.length > 0);

    if (personalities.length === 0) {
      return 'You are a friendly bot'; // Fallback if no personalities are found
  }
  const randomIndex = Math.floor(Math.random() * personalities.length);
  return personalities[randomIndex];
}

class LLMBot {
  constructor() {
    // Instantiate the OpenAI client with the config options we just read
    this.llm = new OpenAI(llmOptions);

    // Prompt fragments we’ll reuse
    this.writePrompt =
      `You are playing a collaborative storytelling game.  
Continue the following story with exactly ONE short, creative sentence.  
Do NOT repeat what’s already written.`;

    this.votePrompt =
      `You are playing a collaborative storytelling game.  
Below is the story so far, followed by a list of possible next sentences.  
Reply with ONLY the number (0-indexed) of the sentence you like best.`;

    this.banterPrompt =
      `You are a playful, slightly sarcastic AI taking part in a writing-game chat.  
Keep it short, fun, and on-topic.`;

    this.personality = generateBotPersonality();
  }

  async Write(story_so_far) {
    const msg = `${this.personality}. ${this.writePrompt}\n\nStory so far:\n${story_so_far}`;
    const response = await this.llm.chat.completions.create({
      model: llmOptions.model || 'meta-llama/Llama-3.2-3B-Instruct-Turbo',
      messages: [{ role: 'user', content: msg }],
      temperature: llmOptions.temperature ?? 0.8,
      max_tokens: llmOptions.max_tokens ?? 80,
    });
    //console.log(`Received response from LLM: ${response.choices[0].message.content}`);
    return response.choices[0].message.content.trim();
  }

  async Vote(story_so_far, choices) {
    const choicesBlock = choices.map((c, i) => `${i}: ${(c.entry_text)}`).join('\n');
    const msg = `${this.personality}. ${this.votePrompt}\n\nStory:\n${story_so_far}\n\nChoices:\n${choicesBlock}`;
    //console.log(msg);
    const response = await this.llm.chat.completions.create({
      model: llmOptions.model || 'meta-llama/Llama-3.2-3B-Instruct-Turbo',
      messages: [{ role: 'user', content: msg }],
      temperature: 0.2,
      max_tokens: 5,
    });
    const reply = response.choices[0].message.content.trim();
    const match = reply.match(/\d+/);
    return match ? parseInt(match[0], 10) : 0;
  }

  async Banter(chat_so_far) {
    const msg = `${this.banterPrompt}\n\nChat so far:\n${chat_so_far}`;
    const response = await this.llm.chat.completions.create({
      model: llmOptions.model || 'meta-llama/Llama-3.2-3B-Instruct-Turbo',
      messages: [{ role: 'user', content: msg }],
      temperature: llmOptions.temperature ?? 0.8,
      max_tokens: llmOptions.max_tokens ?? 60,
    });
    return response.choices[0].message.content.trim();
  }
}

module.exports = LLMBot;