@tozil/sdk

Know what every AI user costs you. Track AI costs per user, per model, per endpoint — with 2 lines of code.

Install

npm install @tozil/sdk

Quick Start

import tozil from "@tozil/sdk";

tozil.init(); // reads TOZIL_API_KEY from env

That's it. Tozil automatically patches the Anthropic and OpenAI SDKs to track every API call — tokens, latency, model, and cost.

Setup

1. Sign up at tozil.dev and grab your API key
2. Set your environment variable:

TOZIL_API_KEY=tz_your_key_here

3. Call tozil.init() before creating any Anthropic/OpenAI clients:

import tozil from "@tozil/sdk";
import Anthropic from "@anthropic-ai/sdk";

tozil.init();

const client = new Anthropic();
const response = await client.messages.create({
  model: "claude-sonnet-4-20250514",
  max_tokens: 1024,
  messages: [{ role: "user", content: "Hello" }],
});
// ^ This call is automatically tracked

Track Users & Endpoints

Use the Express middleware to automatically tag requests with user and endpoint info:

import tozil from "@tozil/sdk";
import { tozilMiddleware } from "@tozil/sdk/express";
import express from "express";

tozil.init();

const app = express();
app.use(tozilMiddleware({
  getUserId: (req) => req.headers["x-user-id"] as string,
}));

Or set context manually anywhere in your code:

import { withContext } from "@tozil/sdk";

await withContext({ userId: "user_123", endpoint: "/chat" }, async () => {
  // All AI calls inside here are tagged with this user & endpoint
  await client.messages.create({ ... });
});

Options

tozil.init({
  apiKey: "tz_...",         // default: process.env.TOZIL_API_KEY
  baseUrl: "https://...",   // default: https://app.tozil.dev/api/v1
  flushInterval: 5000,      // ms between flushes (default: 5000)
  maxBatchSize: 100,         // events per batch (default: 100)
  debug: false,              // log tracking events to console
});

How It Works

• Monkey-patches Anthropic and OpenAI SDK prototypes at init time
• Uses Node.js AsyncLocalStorage for zero-config context propagation
• Buffers events in memory, flushes every 5s (or at batch size)
• Silent failure — never throws, never blocks your critical path
• Zero runtime dependencies

Requirements

• Node.js >= 18
• @anthropic-ai/sdk and/or openai as peer dependencies (install whichever you use)

License

MIT