LLM Spend TripwireChecklistCodeHosted waitlist

Get a Slack or Discord alert when your Claude API usage spikes

LLM Spend Tripwire is a free, open-source Cloudflare Worker that you run in your own account. Every 5 minutes it reads your Anthropic organization's usage by API key and model, and alerts you when a key suddenly uses far more than usual, when a key or model you haven't seen appears, or when keys are busy outside your working hours.

Code: repo opening shortlyRead the quick checklist

Self-hosted with your own Anthropic Admin key. Your usage data goes only to your own Slack or Discord channel, never to us.

Why now

Stolen AI keys run up big bills fast. On November 2, 2026, Google Threat Intelligence Group and Mandiant reported "LLMjacking" incidents with "fraudulent charges associated with resource consumption exceeding $100,000 USD in a single day, with cumulative charges at some organizations reaching $1 million USD before access was revoked." Their advice includes using "spending caps to set budgets and avoid over-spending."

Keys leak through other services too. On November 10, 2026, The Hacker News reported that Fireworks AI found its breach was wider than first disclosed. Its first notice said an intruder using internal Fireworks credentials captured environment variables from evaluation jobs in some customer accounts, including API keys and tokens stored as secrets. Fireworks said the evidence links the attackers to a group "aimed at selling tokens from unauthorized access to large language models (LLMs) to illicit routing gateways."

A spending cap stops the bleeding at a limit. Tripwire is meant to tell you sooner, while the bill is still small.

How it works

  1. You deploy the Worker to your Cloudflare account (one button or wrangler deploy), then add your Anthropic Admin key and a Slack or Discord webhook URL as secrets.
  2. Every 5 minutes it reads your usage from Anthropic's Usage and Cost API in 1-minute buckets, grouped by API key and model, and lists your active API keys.
  3. It compares each key with its own recent average and posts to your channel when:
    • a key's busiest minute is more than 5x its 6-hour average and above 50,000 tokens (both adjustable);
    • a new API key or model appears;
    • keys are used outside the working hours and time zone you set.

Each alert includes an estimated cost (tokens x Anthropic's list price). It's an estimate, not your bill: Anthropic's cost data is daily only, and its usage data typically appears within 5 minutes. Example alert (made-up data):

LLM Spend Tripwire: 2 alerts for your Anthropic organization
- New API key: key "temp-test" (sk-ant-api03-Ef3...vW7X) is active, created 2026-11-16T15:18:40Z.
- Spike: key "prod-backend" (sk-ant-api03-Ab1...xY9Z) used 481,500 tokens in the minute starting 2026-11-16 15:24 UTC (11.6x its average of 41,500/min over the last 6h). Estimated cost of that minute: about $3.25.

Costs are estimates (tokens x list price), not your bill.

Who it's for

Developers and small teams with Claude API keys in production who want an early warning without buying an observability platform.

You need a Claude Console organization. Anthropic's Admin API "isn't available to organizations on the claude.ai Team plan or to individual Console accounts." An individual Console account must first set up an organization on the Organization page in the Claude Console. Claude Enterprise organizations and Claude Platform on AWS use different APIs and aren't supported.

About the Admin key: Console Admin keys can't be limited to read-only. Anthropic's docs say "every key has full access to the endpoints that accept it." Tripwire only makes read requests with it, and its code is short and public so you can check. Keep the key in a Worker secret, and delete it in the Console if you stop using Tripwire.

LLMjacking quick checklist

Steps you can take in the Claude Console now, with or without Tripwire. Each one links to the Anthropic doc it comes from.

  1. Look at recent usage. Open the Usage page in the Claude Console and look for spikes, models you don't use, or activity at odd hours. For a per-key breakdown, Anthropic's Usage API can group usage by API key (it needs an Admin key).
  2. Review every API key. On the API keys page, disable or delete any key you don't recognize or think may have leaked. Anthropic notes that Disable can be undone and Delete is permanent.
  3. Rotate keys that may have been exposed. Create a new key, switch your apps to it, then delete the old one. Anthropic recommends you "rotate them periodically, and disable or delete any key you suspect has leaked" (source).
  4. Set a monthly spend limit. On the Spend limits page in the Console, set your own limit below your tier's cap. You can also set limits per workspace. When a limit is reached, API requests fail until access resumes, so pick a number you can live with (source).
  5. Give new keys an expiration date, so a leaked key stops working on its own (source).
  6. Keep keys out of code and chat. Anthropic recommends storing keys in a secrets manager (source). Google's report recommends automated secret scanning of your code repositories (source).
  7. Consider IP allowlisting (beta). It limits where your keys work. It also applies to the Admin API, so allowlisting can block tools that call it from changing IPs, including Tripwire on Cloudflare (source).
  8. Check who has Admin keys. Only members with the admin role can create them, and each one has full admin access. Keep as few as you need (source).
  9. Turn on an alert. Use Tripwire, or one of the partner integrations Anthropic lists on its Usage and Cost API page.

What it can't do

Don't want to host it yourself?

Join the waitlist for the hosted version (planned $9/month).

No launch date yet, and joining doesn't commit you to anything. We'll use your email only to tell you about the hosted version. See Privacy.