@stax/sdk

Drop-in usage metering for Node apps. One init() and every outbound API call your app makes (Claude, OpenAI, Firecrawl, Exa, …) is captured and reported to Stax — no per-vendor wrappers, no proxy, no card data.

Setup

Option A — scaffold (recommended)

npm install @stax/sdk
npx @stax/sdk init --key stax_yourkey --endpoint https://app.stax.dev

This drops a stax/ folder into your app (like trigger.dev init):

stax/
├─ config.ts   # your key, endpoint & options (defineConfig)
└─ index.ts    # boots Stax — import this once

Then load it once — the only edit to your code:

import './stax';               // first line of your startup file

…or preload it without touching your code at all:

node --import ./stax/index.mjs your-app.js

config.ts is fully typed via defineConfig, so you get autocomplete on every option.

Option B — manual

import * as Stax from '@stax/sdk';

Stax.init({
  apiKey: process.env.STAX_KEY!,        // identifies your project
  endpoint: 'https://app.stax.dev',     // optional; defaults to local dev
});

Either way: call once, as early as possible. That's it — your existing vendor SDK calls are now metered automatically.

Attribute usage to your end-users (optional)

await Stax.withTenant('customer-123', async () => {
  await anthropic.messages.create({ ... });   // tagged to customer-123
});

Wrap a request/job once; every captured call inside it — however deep — is tagged to that tenant via AsyncLocalStorage.

How it works

• Hooks global fetch (undici) and http/https core, so it captures calls regardless of which HTTP library a vendor SDK uses.
• For vendors that report usage (Anthropic/OpenAI token counts), it reads the exact number from the response; otherwise Stax estimates per call.
• Events are batched and shipped in the background using the original fetch — if Stax is unreachable it never blocks or breaks your app.

Scope

It meters traffic from the process where it's installed (your server-side egress). It does not see calls made outside your app (e.g. a vendor's web UI).