Token Counter API

Know exactly how many tokens your text uses before calling an LLM. Supports GPT-4o, Claude, Gemini, and more. One POST request. No library installs.

The problem it solves

Context overflows kill agent runs at the worst moment. Counting tokens correctly means knowing when to summarize, chunk, or truncate — before the call fails.

How it works

POST /api/tools/token-counter

{
  "text": "Analyze this document and
           extract key action items...",
  "model": "gpt-4o"
}

{
  "tokens": 14,
  "model": "gpt-4o",
  "limit": 128000,
  "remaining": 127986,
  "percentUsed": 0.01
}

Common use cases

TypeScript SDK

import { AgentToolbelt } from "agent-toolbelt";

const toolbelt = new AgentToolbelt({ apiKey: process.env.TOOLBELT_KEY });

async function prepareContext(history: Message[], doc: string) {
  const { percentUsed, remaining } = await toolbelt.tokenCounter({
    text: buildContext(history, doc),
    model: "gpt-4o",
  });

  if (percentUsed > 80) {
    return summarizeHistory(history, remaining);
  }

  return buildContext(history, doc);
}