Add 2 lines of code. Wait 48 hours. Get a waste audit showing exactly how much money you're burning on Anthropic API calls — with specific fixes and dollar amounts.
// Before: 0 lines changed
import Anthropic from "@anthropic-ai/sdk";
// After: 2 lines added
import Anthropic from "@anthropic-ai/sdk";
+ import { costly } from "costly";
- const anthropic = new Anthropic();
+ const anthropic = costly({ projectId, apiKey }).wrap(new Anthropic());
// Everything else stays identical
const msg = await anthropic.messages.create({ ... });Install. Wait. Save money. No dashboards to learn.
npm install costly. Add 2 lines of code. Deploy.
Costly silently logs your API calls. Zero latency impact.
Get an email: "Costly found $127/mo in recoverable waste."
Each finding shows what's wrong, how to fix it, and how much you'll save.
Each finding comes with a dollar amount and specific fix instructions.
Repeated system prompts burning input tokens on every call
Using Sonnet/Opus for tasks Haiku handles fine
Identical prompts sent hundreds of times
One feature silently consuming 70%+ of budget
Failed calls that still bill for input tokens
max_tokens set way higher than actual output
Spend growing faster than usage
Most developers waste 30-60% of their Anthropic bill on the wrong model, duplicate calls, or bloated prompts. Find out in 48 hours.
Open-source SDK. Free waste audit during beta.