Toka is an AI Cost Optimizer SDK for developers to track token usage, estimate costs in real-time, reduce API spend, and optimize AI model usage. Save money, reduce redundant calls, and gain full visibility into your AI workloads.

Installation

npm install toka-sdk

Quick Start

import { TokaClient } from 'toka';

Initialize the client

const client = new TokaClient({
  apiKey: process.env.TOKA_API_KEY,
  models: ['gpt-4', 'gpt-4o-mini', 'gpt-3.5'],
  maxCostPerRequest: 0.05,        // Maximum allowed cost per request
  cache: true                     // Enable built-in caching (memory or Redis)
});

Make an API call with cost tracking

const response = await client.chat({
 messages: [{ role: 'user', content: 'Hello world' }]
});

console.log(`Text: ${response.text}`);
console.log(`Cost: $${response.cost}`);
console.log(`Tokens used: ${response.tokens}`);
console.log(`Model used: ${response.modelUsed}`);
console.log(`Cache hit: ${response.cacheHit}`);

Features

•	Token Usage Tracking – Automatically track tokens for every request.
•	Cost Estimation – Real-time cost estimation per request based on model usage.
•	Cost Optimization – Intelligent fallback to cheaper models if cost exceeds budget.
•	Caching – Built-in memory or optional Redis caching to reduce redundant API calls.
•	Multi-Model Support – Automatic switching between multiple AI models.
•	Logging & Analytics – Detailed logs for token usage, cost, cache hits, and fallback events.
•	Optional Enhancements – CLI tool, Dashboard middleware, and Event Hooks for monitoring.

Optional Features

•	Redis Caching – Scalable caching for high-volume usage.
•	Dashboard Middleware – Mountable Express middleware for monitoring AI usage and cost.
•	CLI Tool – Interact with AI APIs directly from your terminal.
•	Event Hooks – Listen to cache hits, fallback events, and request completions.

Testing

Run unit tests for all modules:

npm test

Contributing

We welcome contributions! Please fork the repo and submit pull requests. For guidelines, see CONTRIBUTING.md.

License

MIT License © 2026 Abreham Wondimu Shiferaw