Documentation/Core Concepts/Agentic Tool Call Metering

Agentic Tool Call Metering

Autonomous AI agents do more than stream text—they search the web, scrape websites, and run Python sandboxes.

The Multi-Step Agent Problem

A single user request might use $0.001 in LLM tokens, but execute:

✓1 Google Search ($0.010)
✓1 Web Scraper ($0.005)
✓1 Python Sandbox execution ($0.020)

Total real cost: $0.036. If you only meter the LLM tokens, you lose money on the tools.

Using vibezcheck.session()

Create a scoped customer session to bill prompts and tools together into one customer balance:

typescript

import { generateText, tool } from 'ai';
import { vibezcheck } from 'vibezcheck';
import { z } from 'zod';
export async function POST(req: Request) {
  const { prompt, customer = 'alex@company.com' } = await req.json();
  // ⚡ 1. Create a unified customer session
  const session = vibezcheck.session({ customer });
  const result = await generateText({
    model: session.model('openai/gpt--mini'),
    tools: {
      searchWeb: tool({
        description: 'Live Google Web Search',
        parameters: z.object({ query: z.string() }),
        execute: async ({ query }) => {
          // ⚡ 2. Bill external tool cost ($0.01) into the same balance
          await session.trackTool('google_search', { costUSD: 0.01 });
          return `Results for: ${query}`;
        },
      }),
    },
    prompt,
  });
  return Response.json(result);
}

Core Concepts

Prepaid vs. Postpaid Billing

Next Guide

Core Concepts

Thinking & Reasoning Tokens

💡 Tip: Use Left / Right arrow keys to navigate guidesPage 8 of 15