Limits slip under real conditions
Spend caps that hold in a demo leak during database outages, traffic bursts, restarts and long responses. You find out on the invoice.
For companies that bill their customers for AI usage
Know what every customer costs you in AI, and enforce it.
Stipend tracks the exact AI cost and margin of every customer across every provider, and holds their limits when outages, bursts and long answers would let spend slip through.
The problem
Cost per customer is invisible
Provider bills show totals by model, not by customer. Without per-customer cost you can't see which accounts lose money or price your plans with confidence.
Free tiers get abused
Scripts, shared accounts and sign-up farms turn a generous free tier into an open tab, and nothing flags it until the spend is gone.
What we found
We stress-tested two popular spend-limit tools with real API calls against small per-customer budgets.
3×
budget spent, never blocked
One tool, cut off from its backend while a customer was under budget, never blocked them. Spending reached 3x their limit before our own safety cap ended the test.
62%
of recorded spend lost in an outage
One tool's default setup permanently lost 62% of a customer's recorded spend during a database outage, then handed that budget back after a restart.
+117%
past budget at the default refresh
One tool's limits slipped 117% past budget at its default refresh interval, versus 14.5% at a 10-second refresh.
Tested in October 2026. Full method, versions, and the settings that fix each issue are in our write-up.
What Stipend does
Per-customer limits across providers
One budget per customer covering OpenAI, Anthropic, Google and the rest, enforced before the call is made and kept consistent through outages and restarts.
Exact cost and margin per customer
Every call priced to the token and tied to the customer who made it, next to what you bill them, so you can see margin by account, plan and feature.
Abuse detection
Spots free-tier farming, shared keys and runaway automations from usage patterns, and flags or throttles them before they cost you.
You choose what happens near a limit
Warn, throttle, switch to a cheaper model, require an upgrade, or hard-stop. Set it per plan or per customer.
Become a design partner.
We're working with a small group of companies that bill for AI usage. Design partners get early access, direct input on the roadmap, and preferential pricing.
Check your own limits in two minutes
stipend-check is a free, open-source command-line tool. Point it at your AI gateway or metering SDK and it reports whether your budgets actually enforce, spending at most $0.05.
pip install "git+https://github.com/petreli12/stipend-check.git#subdirectory=tools/stipend-check"
stipend-check --help