Optimizing AI Agent Costs and Tokens

// Example of token estimation for a prompt
const prompt = "Explain how to use React hooks in a component.";
const inputTokens = prompt.split(" ").length;
console.log(`Input tokens: ${inputTokens}`); // Output: Input tokens: 10

// Example of token estimation for a response
const response = "React hooks allow you to use state and other React features without writing a class.";
const outputTokens = response.split(" ").length;
console.log(`Output tokens: ${outputTokens}`); // Output: Output tokens: 14

const aiClient = new AIAssistant({ apiKey: process.env.API_KEY });

// Example of monitoring token usage
const response = await aiClient.complete({
  prompt: "Explain how to optimize token usage in AI development.",
  maxTokens: 150
});

console.log("Input Tokens:", response.usage.promptTokens);
console.log("Output Tokens:", response.usage.completionTokens);
console.log("Total Tokens:", response.usage.totalTokens);

const input = "...very long context string...";
const truncated = input.slice(0, 1024);

from transformers import pipeline

summarizer = pipeline("summarization", model="facebook/bart-large-cnn")
summary = summarizer(context, max_length=256, min_length=50, do_sample=False)

const promptCache = new PromptCache();

// First query
let response1 = await aiAgent.generate({
  prompt: "Explain quantum computing",
  cacheKey: "quantum_explained"
});

// Second query with same prompt
let response2 = await aiAgent.generate({
  prompt: "Explain quantum computing",
  cacheKey: "quantum_explained"
});

// response2 is served from cache
console.log(response1 === response2); // true

const batchedRequests = [
  { prompt: "Summarize this code:", code: "function add(a, b) { return a + b; }" },
  { prompt: "Explain the logic:", code: "if (x > 0) { console.log('positive'); }" }
];

const response = await fetch('/api/batch', {
  method: 'POST',
  body: JSON.stringify({ requests: batchedRequests }),
  headers: { 'Content-Type': 'application/json' }
});

ai:
  cost:
    budget:
      daily: 10.00
      weekly: 50.00
      monthly: 200.00
    alerts:
      email: user@example.com
      threshold:
        daily: 0.8
        weekly: 0.9

const anthropic = require("@anthropic-ai/sdk");
const client = new anthropic.Anthropic();

async function processTask() {
  const response = await client.messages.create({
    model: "claude-3-haiku-20240307",
    max_tokens: 100,
    messages: [
      {
        role: "user",
        content: "Summarize this code in 3 lines."
      }
    ]
  });
  console.log(response.content[0].text);
}

processTask();

// Sample parsed AI cost report structure
const aiCostReport = {
  totalTokens: 1250000,
  totalCost: 45.75,
  modelUsage: {
    gpt_4: { tokens: 800000, cost: 32.00 },
    gpt_3_5_turbo: { tokens: 450000, cost: 13.75 }
  },
  monthlyBreakdown: [
    { month: '2026-01', tokens: 450000, cost: 16.25 },
    { month: '2026-02', tokens: 400000, cost: 15.00 },
    { month: '2026-03', tokens: 400000, cost: 14.50 }
  ]
};