managarten/chat/apps/mobile/services/openai.ts

/**
 * Chat Service - AI API calls via Backend
 * This service wraps the backend API for AI completions
 */
import { availableModels } from '../config/azure';
import { chatApi, modelApi, usageApi, type ChatMessage, type TokenUsage } from './api';

// Re-export types for backward compatibility
export type { ChatMessage };

// Re-export TokenUsage
export type { TokenUsage };

// Chat response type (kept for compatibility)
export type ChatResponse = {
  id: string;
  choices: {
    content_filter_results?: any;
    finish_reason: string;
    index: number;
    logprobs: any;
    message?: {
      content: string;
      refusal?: any;
      role: string;
    };
  }[];
  created: number;
  model: string;
  object: string;
  prompt_filter_results?: any[];
  system_fingerprint?: string;
  usage: {
    prompt_tokens: number;
    completion_tokens: number;
    total_tokens: number;
    completion_tokens_details?: any;
    prompt_tokens_details?: any;
  };
};

// Return type for chat request
export type ChatRequestResult = {
  content: string;
  usage: TokenUsage;
};

// Logging configuration
console.log('Chat Service Konfiguration:', {
  backendUrl: process.env.EXPO_PUBLIC_BACKEND_URL || 'http://localhost:3001',
  availableModels: availableModels.length,
});

/**
 * Calculates estimated cost for an LLM request
 */
export async function calculateTokenCost(
  promptTokens: number,
  completionTokens: number,
  modelId: string
): Promise<number> {
  try {
    // Get cost settings from model
    const modelData = await modelApi.getModel(modelId);

    if (!modelData || !modelData.costSettings) {
      console.warn('Fehler beim Laden der Kosteninformationen, verwende Standardwerte');
      const promptCost = 0.0001;
      const completionCost = 0.0002;

      const cost = (promptTokens * promptCost + completionTokens * completionCost) / 1000;
      return Number(cost.toFixed(6));
    }

    const promptCost = modelData.costSettings.prompt_per_1k_tokens || 0.0001;
    const completionCost = modelData.costSettings.completion_per_1k_tokens || 0.0002;

    const cost = (promptTokens * promptCost + completionTokens * completionCost) / 1000;
    return Number(cost.toFixed(6));
  } catch (error) {
    console.error('Fehler bei der Kostenberechnung:', error);
    return Number(((promptTokens * 0.0001 + completionTokens * 0.0002) / 1000).toFixed(6));
  }
}

/**
 * Logs token usage to the database
 */
export async function logTokenUsage(
  usage: TokenUsage,
  conversationId: string,
  messageId: string,
  userId: string,
  modelId: string
): Promise<void> {
  try {
    const estimatedCost = await calculateTokenCost(
      usage.prompt_tokens,
      usage.completion_tokens,
      modelId
    );

    const success = await usageApi.logTokenUsage({
      conversationId,
      messageId,
      modelId,
      promptTokens: usage.prompt_tokens,
      completionTokens: usage.completion_tokens,
      totalTokens: usage.total_tokens,
      estimatedCost,
    });

    if (success) {
      console.log('Token-Nutzung erfolgreich gespeichert:', {
        conversationId,
        messageId,
        totalTokens: usage.total_tokens,
        estimatedCost,
      });
    } else {
      console.error('Fehler beim Speichern der Token-Nutzung');
    }
  } catch (error) {
    console.error('Fehler beim Loggen der Token-Nutzung:', error);
  }
}

/**
 * Sends a chat request via the backend
 */
export async function sendChatRequest(
  messages: ChatMessage[],
  temperature: number = 0.7,
  maxTokens: number = 800
): Promise<string | ChatRequestResult> {
  console.log('sendChatRequest gestartet mit:', {
    messagesCount: messages.length,
    maxTokens,
  });

  try {
    // Find model deployment from system message
    let modelId = '550e8400-e29b-41d4-a716-446655440000'; // Default to GPT-O3-Mini

    const systemMessage = messages.find(
      (msg) => msg.role === 'system' && msg.content.startsWith('MODEL:')
    );
    if (systemMessage) {
      const deployment = systemMessage.content.split(':')[1].trim();
      console.log('Modell in system Nachricht erkannt:', deployment);

      // Map deployment to model ID
      const deploymentToModelId: Record<string, string> = {
        'gpt-o3-mini-se': '550e8400-e29b-41d4-a716-446655440000',
        'gpt-4o-mini-se': '550e8400-e29b-41d4-a716-446655440004',
        'gpt-4o-se': '550e8400-e29b-41d4-a716-446655440005',
      };

      modelId = deploymentToModelId[deployment] || modelId;
    } else {
      console.warn('Keine System-Nachricht mit MODEL-Präfix gefunden!');
    }

    console.log('Verwende Model ID:', modelId);

    // Filter out MODEL: system messages before sending to API
    const filteredMessages = messages.filter(
      (msg) => !(msg.role === 'system' && msg.content.startsWith('MODEL:'))
    );

    // Send request to backend
    const result = await chatApi.createCompletion({
      messages: filteredMessages,
      modelId,
      temperature,
      maxTokens,
    });

    if (!result) {
      return 'Es tut mir leid, aber ich konnte keine Antwort generieren. Bitte stelle sicher, dass das Backend läuft.';
    }

    return {
      content: result.content,
      usage: result.usage,
    };
  } catch (error) {
    console.error('Fehler bei der Chat-Anfrage:', error);

    if (error instanceof Error) {
      console.error('Fehlerdetails:', {
        name: error.name,
        message: error.message,
        stack: error.stack,
      });
    }

    return `Es tut mir leid, aber ich konnte keine Antwort generieren. Bitte stelle sicher, dass das Backend läuft. Fehlerdetails: ${error instanceof Error ? error.message : 'Unbekannter Fehler'}`;
  }
}