librechat-analytics/services/analytics.js

const { getModelPricing } = require('../config/pricing');

// Cache agents collection in memory, refresh every 5 min
let agentsCache = new Map();
let agentsCacheTime = 0;
const CACHE_TTL = 5 * 60 * 1000;

async function refreshAgentsCache(db) {
  if (Date.now() - agentsCacheTime < CACHE_TTL && agentsCache.size > 0) return;
  try {
    const agents = await db.collection('agents').find({}).toArray();
    agentsCache = new Map();
    for (const a of agents) {
      agentsCache.set(a.id, { name: a.name, model: a.model, provider: a.provider });
    }
    agentsCacheTime = Date.now();
  } catch (e) {
    console.error('Failed to refresh agents cache:', e.message);
  }
}

function resolveModel(model) {
  if (model && model.startsWith('agent_')) {
    const agent = agentsCache.get(model);
    return agent ? agent.model : model;
  }
  return model;
}

function getDateRange(query) {
  const { period, start, end } = query;
  const now = new Date();
  let startDate, endDate;

  if (period === 'custom' && start && end) {
    startDate = new Date(start);
    endDate = new Date(end);
    endDate.setHours(23, 59, 59, 999);
  } else {
    endDate = now;
    switch (period) {
      case '7d': startDate = new Date(now - 7 * 86400000); break;
      case '30d': startDate = new Date(now - 30 * 86400000); break;
      case '24h':
      default: startDate = new Date(now - 24 * 3600000); break;
    }
  }
  return { startDate, endDate };
}

async function getSummary(db, query) {
  await refreshAgentsCache(db);
  const { startDate, endDate } = getDateRange(query);

  const [tokenResult, userCount, convCount] = await Promise.all([
    db.collection('transactions').aggregate([
      { $match: { createdAt: { $gte: startDate, $lte: endDate } } },
      {
        $group: {
          _id: null,
          totalTokens: { $sum: { $abs: '$rawAmount' } },
          totalCost: { $sum: { $abs: '$tokenValue' } },
        }
      }
    ]).toArray(),
    db.collection('transactions').distinct('user', {
      createdAt: { $gte: startDate, $lte: endDate }
    }),
    db.collection('transactions').distinct('conversationId', {
      createdAt: { $gte: startDate, $lte: endDate }
    }),
  ]);

  const t = tokenResult[0] || { totalTokens: 0, totalCost: 0 };
  return {
    totalTokens: t.totalTokens,
    totalCost: t.totalCost / 1_000_000,
    activeUsers: userCount.length,
    conversations: convCount.length,
  };
}

async function getTopUsers(db, query, limit = 10) {
  const { startDate, endDate } = getDateRange(query);

  const results = await db.collection('transactions').aggregate([
    { $match: { createdAt: { $gte: startDate, $lte: endDate } } },
    {
      $group: {
        _id: '$user',
        totalTokens: { $sum: { $abs: '$rawAmount' } },
        totalCost: { $sum: { $abs: '$tokenValue' } },
        conversations: { $addToSet: '$conversationId' },
      }
    },
    { $sort: { totalCost: -1 } },
    { $limit: limit },
    {
      $lookup: {
        from: 'users',
        localField: '_id',
        foreignField: '_id',
        as: 'userInfo'
      }
    },
    { $unwind: { path: '$userInfo', preserveNullAndEmptyArrays: true } },
    {
      $project: {
        name: { $ifNull: ['$userInfo.name', 'Unknown'] },
        email: { $ifNull: ['$userInfo.email', ''] },
        totalTokens: 1,
        totalCost: { $divide: ['$totalCost', 1_000_000] },
        conversationCount: { $size: '$conversations' },
      }
    }
  ]).toArray();

  return results;
}

async function getTopModels(db, query, limit = 10) {
  await refreshAgentsCache(db);
  const { startDate, endDate } = getDateRange(query);

  const raw = await db.collection('transactions').aggregate([
    { $match: { createdAt: { $gte: startDate, $lte: endDate } } },
    {
      $group: {
        _id: { model: '$model', tokenType: '$tokenType' },
        totalTokens: { $sum: { $abs: '$rawAmount' } },
        totalCost: { $sum: { $abs: '$tokenValue' } },
      }
    }
  ]).toArray();

  // Resolve agents to underlying LLM and re-aggregate
  const modelMap = new Map();
  for (const r of raw) {
    const resolvedModel = resolveModel(r._id.model);
    const key = `${resolvedModel}::${r._id.tokenType}`;
    if (!modelMap.has(key)) {
      modelMap.set(key, { model: resolvedModel, tokenType: r._id.tokenType, totalTokens: 0, totalCost: 0 });
    }
    const entry = modelMap.get(key);
    entry.totalTokens += r.totalTokens;
    entry.totalCost += r.totalCost;
  }

  // Pivot into per-model summary
  const models = new Map();
  for (const entry of modelMap.values()) {
    if (!models.has(entry.model)) {
      models.set(entry.model, { model: entry.model, promptTokens: 0, completionTokens: 0, promptCost: 0, completionCost: 0, totalCost: 0 });
    }
    const m = models.get(entry.model);
    if (entry.tokenType === 'prompt') {
      m.promptTokens += entry.totalTokens;
      m.promptCost += entry.totalCost / 1_000_000;
    } else {
      m.completionTokens += entry.totalTokens;
      m.completionCost += entry.totalCost / 1_000_000;
    }
    m.totalCost = m.promptCost + m.completionCost;
  }

  return Array.from(models.values())
    .sort((a, b) => b.totalCost - a.totalCost)
    .slice(0, limit);
}

async function getTopAgents(db, query, limit = 10) {
  await refreshAgentsCache(db);
  const { startDate, endDate } = getDateRange(query);

  const results = await db.collection('transactions').aggregate([
    { $match: { createdAt: { $gte: startDate, $lte: endDate }, model: { $regex: /^agent_/ } } },
    {
      $group: {
        _id: '$model',
        totalTokens: { $sum: { $abs: '$rawAmount' } },
        totalCost: { $sum: { $abs: '$tokenValue' } },
        conversations: { $addToSet: '$conversationId' },
      }
    },
    { $sort: { totalCost: -1 } },
    { $limit: limit },
    {
      $project: {
        agentId: '$_id',
        totalTokens: 1,
        totalCost: { $divide: ['$totalCost', 1_000_000] },
        conversationCount: { $size: '$conversations' },
      }
    }
  ]).toArray();

  return results.map(r => {
    const agent = agentsCache.get(r.agentId);
    return {
      ...r,
      agentName: agent ? agent.name : r.agentId,
      underlyingModel: agent ? agent.model : 'Unknown',
      provider: agent ? agent.provider : 'Unknown',
    };
  });
}

async function getCostBreakdown(db, query) {
  await refreshAgentsCache(db);
  const { startDate, endDate } = getDateRange(query);

  const raw = await db.collection('transactions').aggregate([
    { $match: { createdAt: { $gte: startDate, $lte: endDate } } },
    {
      $group: {
        _id: { model: '$model', tokenType: '$tokenType' },
        totalCost: { $sum: { $abs: '$tokenValue' } },
      }
    }
  ]).toArray();

  const models = new Map();
  for (const r of raw) {
    const resolved = resolveModel(r._id.model);
    if (!models.has(resolved)) {
      models.set(resolved, { model: resolved, inputCost: 0, outputCost: 0 });
    }
    const m = models.get(resolved);
    if (r._id.tokenType === 'prompt') {
      m.inputCost += r.totalCost / 1_000_000;
    } else {
      m.outputCost += r.totalCost / 1_000_000;
    }
  }

  return Array.from(models.values())
    .map(m => ({ ...m, totalCost: m.inputCost + m.outputCost }))
    .sort((a, b) => b.totalCost - a.totalCost);
}

async function getUsageOverTime(db, query) {
  const { startDate, endDate } = getDateRange(query);
  const diffMs = endDate - startDate;
  const diffHours = diffMs / 3600000;

  // Use hourly buckets for <=48h, daily for longer
  let dateFormat, bucketLabel;
  if (diffHours <= 48) {
    dateFormat = { $dateToString: { format: '%Y-%m-%dT%H:00', date: '$createdAt' } };
    bucketLabel = 'hour';
  } else {
    dateFormat = { $dateToString: { format: '%Y-%m-%d', date: '$createdAt' } };
    bucketLabel = 'day';
  }

  const results = await db.collection('transactions').aggregate([
    { $match: { createdAt: { $gte: startDate, $lte: endDate } } },
    {
      $group: {
        _id: dateFormat,
        totalTokens: { $sum: { $abs: '$rawAmount' } },
        totalCost: { $sum: { $abs: '$tokenValue' } },
      }
    },
    { $sort: { _id: 1 } },
  ]).toArray();

  return {
    bucketType: bucketLabel,
    data: results.map(r => ({
      time: r._id,
      tokens: r.totalTokens,
      cost: r.totalCost / 1_000_000,
    })),
  };
}

module.exports = {
  getSummary,
  getTopUsers,
  getTopModels,
  getTopAgents,
  getCostBreakdown,
  getUsageOverTime,
};