llm-cost-tracker

Lightweight, zero-dependency LLM API cost & token usage tracker for Node.js / TypeScript.

Track every token and dollar spent across OpenAI, Anthropic (Claude), Google Gemini, Mistral, Groq, and DeepSeek — with a single line of code.

✨ Features

• 🎯 6 Providers — OpenAI, Anthropic, Gemini, Mistral, Groq, DeepSeek
• 💰 35+ Models — Built-in pricing database (April 2026 prices)
• 🔍 Auto-Detection — Automatically detects provider from response shape
• 🧩 Fuzzy Matching — gpt-4o-2024-11-20 matches gpt-4o pricing
• 📊 Rich Summaries — Aggregated stats by provider, model, time period
• 💾 Pluggable Storage — In-memory, JSON file, or bring your own
• 🔌 Middleware — Drop-in fetch wrapper + Vercel AI SDK integration
• 📤 Export — JSON and CSV export
• 🪶 Zero Dependencies — ~22KB bundled, works everywhere
• 🔒 TypeScript-first — Full type safety, ESM + CJS

📦 Installation

npm install llm-cost-tracker

🚀 Quick Start

import { CostTracker } from 'llm-cost-tracker';

const tracker = new CostTracker();

// Track a request
tracker.track('openai', 'gpt-4o', {
  inputTokens: 1000,
  outputTokens: 500,
});

// Get cost summary
const summary = await tracker.summary();
console.log(`Total cost: $${summary.totalCost}`);
// Total cost: $0.0075

📖 Usage

Method 1: Manual Tracking

When you already have token counts from the API response:

const record = tracker.track('openai', 'gpt-4o', {
  inputTokens: 1000,
  outputTokens: 500,
});

console.log(record.totalCost);  // 0.0075
console.log(record.inputCost);  // 0.0025
console.log(record.outputCost); // 0.005

Method 2: Parse Raw API Response

Pass the raw API response object and let the tracker extract tokens automatically:

// OpenAI
const response = await fetch('https://api.openai.com/v1/chat/completions', { ... });
const body = await response.json();
tracker.trackResponse('openai', body);

// Anthropic
tracker.trackResponse('anthropic', anthropicResponse);

// Auto-detect provider from response shape
tracker.trackResponse(anyResponse); // Works for all providers!

Method 3: Wrap Async Calls

Wrap your API call — tracking happens automatically:

const result = await tracker.wrap('openai', () =>
  openai.chat.completions.create({
    model: 'gpt-4o',
    messages: [{ role: 'user', content: 'Hello!' }],
  })
);
// result is the normal API response — tracking happened in the background

🔌 Middleware

Fetch Wrapper

Intercept all LLM API calls automatically:

import { CostTracker } from 'llm-cost-tracker';
import { createTrackedFetch } from 'llm-cost-tracker';

const tracker = new CostTracker();
const trackedFetch = createTrackedFetch(tracker);

// Use as drop-in replacement for fetch
const response = await trackedFetch('https://api.openai.com/v1/chat/completions', {
  method: 'POST',
  headers: { 'Authorization': `Bearer ${apiKey}` },
  body: JSON.stringify({ model: 'gpt-4o', messages }),
});

Vercel AI SDK

import { wrapLanguageModel } from 'ai';
import { openai } from '@ai-sdk/openai';
import { CostTracker } from 'llm-cost-tracker';
import { costTrackerMiddleware } from 'llm-cost-tracker/middleware/vercel-ai';

const tracker = new CostTracker();

const model = wrapLanguageModel({
  model: openai('gpt-4o'),
  middleware: costTrackerMiddleware(tracker),
});

// All calls through this model are now tracked

📊 Querying & Reports

Summary

const summary = await tracker.summary();

console.log(summary.totalCost);         // Total USD spent
console.log(summary.requestCount);      // Number of API calls
console.log(summary.avgCostPerRequest); // Average cost per call
console.log(summary.byProvider);        // Breakdown by provider
console.log(summary.byModel);          // Breakdown by model

Filtered Queries

import { startOfToday, startOfMonth } from 'llm-cost-tracker';

// Today's costs
const today = await tracker.summary({ from: startOfToday() });

// This month's OpenAI costs
const openai = await tracker.summary({
  provider: 'openai',
  from: startOfMonth(),
});

// Last 10 records
const recent = await tracker.records({ limit: 10 });

Export

// Export as JSON
const json = await tracker.export('json');

// Export as CSV
const csv = await tracker.export('csv');

💾 Storage Options

In-Memory (Default)

const tracker = new CostTracker(); // Uses memory by default
const tracker = new CostTracker({ store: 'memory', maxRecords: 50000 });

JSON File

const tracker = new CostTracker({
  store: 'json-file',
  storePath: './costs/llm-usage.jsonl',
});

Custom Adapter

const tracker = new CostTracker({
  store: {
    save: async (record) => { await myDB.insert(record); },
    getAll: async (filter) => { return await myDB.query(filter); },
    clear: async () => { await myDB.truncate(); },
  },
});

💲 Pricing

Built-in Models (35+)

| Provider | Models | |:---|:---| | OpenAI | gpt-4o, gpt-4o-mini, o3, o3-mini, o1, gpt-4-turbo, gpt-3.5-turbo | | Anthropic | claude-opus-4, claude-sonnet-4, claude-haiku-4, claude-3.5-sonnet | | Google Gemini | gemini-2.5-pro, gemini-2.5-flash, gemini-2.5-flash-lite | | Mistral | mistral-large, mistral-medium, mistral-small | | Groq | llama-3.1-8b/70b/405b, mixtral-8x7b, gemma-7b | | DeepSeek | deepseek-v4-flash, deepseek-v4-pro, deepseek-chat |

Custom Pricing

const tracker = new CostTracker({
  customPricing: [
    { provider: 'openai', model: 'gpt-5', inputPer1M: 5.00, outputPer1M: 15.00 },
    { provider: 'my-provider', model: 'my-model', inputPer1M: 0.01, outputPer1M: 0.02 },
  ],
});

// Or add at runtime
tracker.addPricing([
  { provider: 'openai', model: 'gpt-5-turbo', inputPer1M: 3.00, outputPer1M: 10.00 },
]);

🔔 Usage Callbacks

const tracker = new CostTracker({
  onUsage: (record) => {
    console.log(`💰 ${record.provider}/${record.model}: $${record.totalCost}`);
    // Send to your monitoring system, Slack webhook, etc.
  },
});

📐 API Reference

CostTracker

| Method | Description | |:---|:---| | track(provider, model, usage, metadata?, latencyMs?) | Track with known token counts | | trackResponse(provider, response, metadata?, latencyMs?) | Parse raw API response | | trackResponse(response) | Auto-detect provider and parse | | wrap(provider, fn, metadata?) | Wrap async call, auto-track | | summary(filter?) | Get aggregated cost summary | | records(filter?) | Get raw usage records | | lastRecord() | Get the last tracked record | | export('json' \| 'csv', filter?) | Export records | | reset() | Clear all records | | addPricing(entries) | Add custom pricing at runtime | | getPrice(provider, model) | Look up model pricing |

QueryFilter

| Field | Type | Description | |:---|:---|:---| | provider | string | Filter by provider | | model | string | Filter by model | | from | Date | Records after this date | | to | Date | Records before this date | | limit | number | Max records to return |

🤝 Contributing

Contributions are welcome! Please open an issue or submit a pull request.

📄 License

MIT