/**
 * Dynamic model fetching from Berget API
 */

import { getModelsEndpoint } from '../constants';
import { logDebug, logError } from './debug';

// Response from /v1/models/chat endpoint
interface ChatModel {
  aliases?: string[];
  contextWindow: number;
  id: string;
  inputPricePerToken: number;
  lifecycleState?: string;
  outputPricePerToken: number;
}

interface ChatModelsResponse {
  models: ChatModel[];
}

// Models that support vision/image input
export const VISION_MODELS = new Set(['moonshotai/Kimi-K2.6', 'moonshotai/Kimi-K3']);

// Cache for models to avoid repeated API calls
let cachedModels: Record<string, object> | undefined;
let cacheTimestamp = 0;
const CACHE_TTL_MS = 5 * 60 * 1000; // 5 minutes

/**
 * Clears the model cache (useful for testing or forcing refresh)
 */
export function clearModelCache(): void {
  cachedModels = undefined;
  cacheTimestamp = 0;
}

/**
 * Fetches available chat models from Berget API
 * Uses /v1/models/chat which only returns text models (no embeddings, rerankers, whisper)
 */
export async function fetchBergetModels(): Promise<Record<string, object>> {
  // Return cached models if still valid
  if (cachedModels && Date.now() - cacheTimestamp < CACHE_TTL_MS) {
    logDebug('Returning cached models');
    return cachedModels;
  }

  logDebug('Fetching chat models from Berget API');

  try {
    const response = await fetch(getModelsEndpoint(), {
      headers: {
        'Content-Type': 'application/json',
      },
      method: 'GET',
    });

    if (!response.ok) {
      logError(`Failed to fetch models: ${response.status}`);
      return getDefaultModels();
    }

    const data = (await response.json()) as ChatModelsResponse;

    const models: Record<string, object> = {};

    for (const model of data.models) {
      const modelConfig: Record<string, unknown> = {
        contextWindow: model.contextWindow,
        // OpenCode model config
        name: model.id,
      };

      // Enable vision support for models that support image input
      if (VISION_MODELS.has(model.id)) {
        modelConfig.attachment = true;
        modelConfig.modalities = {
          input: ['text', 'image'],
          output: ['text'],
        };
      }

      models[model.id] = modelConfig;
    }

    // Cache the results
    cachedModels = models;
    cacheTimestamp = Date.now();

    logDebug(`Fetched ${Object.keys(models).length} chat models`);

    return models;
  } catch (error) {
    logError('Error fetching models', error);
    return getDefaultModels();
  }
}

/**
 * Fallback models if API is unreachable
 */
function getDefaultModels(): Record<string, object> {
  return {
    'meta-llama/Llama-3.1-8B-Instruct': {},
    'meta-llama/Llama-3.3-70B-Instruct': {},
    'mistralai/Mistral-Small-3.2-24B-Instruct-2506': {},
    'openai/gpt-oss-120b': {},
  };
}