Your agent sends 1,500 tokens of tools before the user says a word

import { countTokens } from 'tokens-usage'
import OpenAI from 'openai'

const client = new OpenAI()
const messages = [{ role: 'user', content: 'Find me a laptop' }]
const tools = [/* your function definitions */]

// 1. Tool schema overhead — compare API usage with and without tools
const [withTools, withoutTools] = await Promise.all([
  client.chat.completions.create({ model: 'gpt-4o', messages, tools, max_tokens: 1 }),
  client.chat.completions.create({ model: 'gpt-4o', messages, max_tokens: 1 }),
])
const schemaOverhead =
  withTools.usage!.prompt_tokens - withoutTools.usage!.prompt_tokens

// 2. Tool history overhead — toggle countAssistantTools on message blocks
const history = [
  { role: 'user', content: 'Find me a laptop' },
  { type: 'function_call', call_id: 'c1', name: 'search', arguments: '{}' },
  { type: 'function_call_output', call_id: 'c1', output: '{"results":[]}' },
]

const withHistory = await countTokens({
  provider: 'openai',
  model: 'gpt-4o',
  content: history,
  countAssistantTools: true,
})

const withoutHistory = await countTokens({
  provider: 'openai',
  model: 'gpt-4o',
  content: history,
  countAssistantTools: false,
})

const historyOverhead = withHistory.tokens - withoutHistory.tokens

console.log(`Schema overhead: ${schemaOverhead} tokens/request`)
console.log(`History overhead: ${historyOverhead} tokens in current messages`)