Refuse the runaway call at the cap. 402, not a surprise invoice.
You resell AI at a fixed price. SixDecimal attributes every token to a client, holds a hard budget cap in the request path, and reads true margin per account — so no client's agent can quietly burn your month.
Now onboarding a handful of design partners
Every call reserves budget atomically before the upstream request — reserve → commit → refund, counted to the micro-dollar.
Live mechanism · example figures
One gate in front of every provider
Between the provider's meter and your client's invoice, margin quietly disappears.
You carry the cost risk. Provider dashboards are read-only and lag by hours; by the time an alert fires, the money is spent — and you still can't say which client burned it.
“I don't know what a client's AI truly costs me until the provider invoice lands at the end of the month.”
“A client's agent can loop, or someone can abuse it, and burn hundreds of dollars in hours — with no real brake.”
“I bill a fixed price or per project, but I genuinely don't know which clients I make money on and which I lose.”
Metering tells you what happened. A gate decides what happens next.
One atomic pipeline on the request path — reserve, commit, refund — with Postgres as the durable source of truth. Not a report you read tomorrow.
Attribute every call
A header (x-sd-customer-id) or a virtual key tags each call to a client, feature and team. No app rewrite — point at the gateway or change one base URL.
Hold a hard cap, inline
Before the upstream call, SixDecimal reserves the estimated cost against every applicable budget. Over the cap? The gate refuses with a fail-closed 402 and the provider is never called. On completion the reservation commits to real cost or refunds.
Cross cost with revenue
Connect Stripe (read-only) and SixDecimal joins attributed cost against real revenue per client, per period — so an account where cost outran revenue is flagged before the quarter closes.
Built for the team that carries the cost between the meter and the invoice.
Agencies & studios
You ship client chatbots, copilots and automations on LLM APIs and invoice a retainer or fixed price. The token cost is yours to absorb.
Product teams reselling AI
AI features are part of your paid product. You need per-customer cost next to per-customer revenue, not one lump provider bill.
Platforms with many payers
You have dozens of billable accounts, each with its own budget and its own way to run up a bill you didn't price for.
We're pre-revenue. So we show the mechanism, not a wall of logos.
No customer logos, no testimonials, no invented usage stats — there aren't any yet, and pretending otherwise is exactly the kind of dishonesty we built this to end. The proof is that the cap is atomic by construction.
Metering ≠ enforcement
402 · refusedEven the popular open-source proxy has open enforcement bugs where budgets simply don't hold under load:
- #26672Concurrent requests bypass the budget check entirely.
- #27735Stale spend: cached totals let overspend slip through.
- #25495Budgets never reset on their window — silently blocking or leaking.
Open LiteLLM issues as of 2026. SixDecimal's ledger is atomic by construction — reserve, commit, refund — and verified under 1,000 concurrent goroutines in its test suite.
Every dollar is counted to the micro-dollar — never rounded on the hot path.
example figure
Put a hard ceiling on every client's spend — before the invoice does it for you.
We're pre-revenue and onboarding a handful of agencies and studios as design partners. Attribute cost per client, cap runaway agents, and see real margin per account.