Built for Next.js AI apps, and for the AI that writes them.
Your AI app works.Your AI bill has no ceiling.
Every AI call costs you money, and nothing stops one user, one bot or one bug from making ten thousand of them. You find out when the invoice arrives.
One bad night · A month of revenue
A limit for every user.A cap on the whole bill.
Mantis checks each AI call before it runs and prices it after, so no user, bot or bug can spend more than you allow.
4,000
Calls from one user, stopped at 200
A limit per user
Give every user a daily allowance of AI calls. One heavy user or one bot stops at the limit, and everyone else carries on.
$5k$
A surprise bill, capped at $50
A hard monthly cap
Set the most your AI may cost this month. At the cap new calls stop, and a kill switch in the dashboard stops them at any time.
30dm
From the month-end invoice to an alert in a minute
Told in a minute
When spend spikes you get an email that names the user and the route, with a prompt your AI can use to fix it.
Tell your AI.It wires the rest.
The Mantis MCP server and skill show Cursor and Claude Code how to add it: wrap each AI call, set the limits and put one key in .env.
Try: “Cap each user at 200 AI calls a day”
Inside your app.Not in front of it.
Mantis is a package in your Next.js server, not a proxy. It asks whether a call may go ahead, then reports what it cost.
Your keys stay yours
OpenAI, Anthropic and Stripe keys never leave your server. Mantis sees counts and costs, never your prompts.
Works with your SDK
The OpenAI SDK, the Anthropic SDK and the Vercel AI SDK, in a Next.js app.
Built for your AI
An MCP server and a skill, so Cursor and Claude Code can install it and fix what an alert points to.
Illustration: three Mantis alerts. A user stopped at their daily limit, the monthly cap reached, and a spend spike traced to one route.
Showing Limit per user: One user made 4,000 calls in an hour
Your AI app deserves
a ceiling.
Mantis is launching soon, starting with Next.js apps.