For Platforms & Resellers
You resell AI by the token. You bill by the seat. The difference is your margin.
Your costs scale with every prompt and completion. Your prices — DAU tiers, seats, flat plans — don't. On every heavy customer you undercharge, on every overage you eat the margin, and at month-end you reconcile by hand. Tokenality closes that gap: meter once, rate with your margin across every provider, reconcile, and bill your customers — without touching how you call models.
The last business day of the month
Maya runs a white-label AI platform. She has 40 tenants, one flat price tier, and no idea which of them is quietly costing her money.
4:30 PM
One tenant is on the “growth” tier, burning enterprise-tier tokens.
Their bracket says $199. Their real consumption is several times that. The margin difference has been leaking every month — and no report shows it.
The switch
She already meters every token. She just wasn't billing for them.
Tokenality tails her existing usage table, rates each tenant with her margin across every provider, and drafts the invoice — no change to how she calls models.
Next close
Every tenant billed for what they actually used.
Reconciliation shows events == invoiced before she closes. The heavy tenant pays their real cost-plus-margin. The leak is gone.
What becomes true
Revenue accuracy you didn't have to build.
Stop eating overages — accurate per-customer usage means you bill every token you serve, not a proxy tier that undercharges your heaviest customers.
Metering accuracy is revenue accuracy — idempotent events and a reconciliation pass prove events-generated equals invoiced before you close the month.
Own your meter — Stripe, your ERP, or a CSV are swappable projections set per customer, never lock-in. Neutral by construction.
Every provider, not just the auto-priced three — Azure, Bedrock, self-hosted, fine-tunes, embeddings, and caching all rate through one card.
The fourth dimension
People. Projects. Budgets. Now — Customers.
Tokenality already meters every token and attributes it across your people, projects, and budgets — the control plane that saves you money internally. The customer dimension is the same meter pointed outward: the tokens you serve to your customers become revenue you rate, reconcile, and bill. Cost on one side of the ledger, revenue on the other. One meter.
# Option A — you already write a usage row per call? Tail it. Zero call-site changes.
tokenality backfill --from-table token_ledger # history rated instantly
# Option B — emit one line after your existing metering hook:
await tk.usage.record({
customerId: tenantId, // YOUR customer of record
model: "claude-sonnet-5", // any provider — we rate them all
inputTokens, outputTokens,
requestId: ledgerRowId, // your own row id = exactly-once, no double-bill
});
# Then: rate with your margin, reconcile, and project the invoice to
# Stripe / your ERP / a CSV — swappable, per customer. Own your meter.Curious what accurate billing would recover for you? The Billing ROI calculator turns your customer count and token volume into a recovered-revenue number in about a minute.
It meters what you already run.
Keep your gateway, your providers, and your Stripe account. Tokenality is the neutral billing layer in between — it turns the tokens you already serve into accurate, reconciled customer revenue.
Ledger adapter
Already write a usage row per call? We tail it on a cursor — zero AI-call-path changes.
Report-mode SDK
One line after your metering hook: tk.usage.record({ customerId, model, tokens })
Every provider
OpenAI, Anthropic, Google, Azure, Bedrock, self-hosted, fine-tunes, embeddings
Rate cards
Markup, flat, or tiered — cost → your price, versioned, config-as-code
Stripe (your account)
We draft the itemized invoice; your Stripe collects. Processor-neutral.
NetSuite / Intacct / QuickBooks
Enterprise AR + ASC-606 rev-rec export (roadmap)
Reconciliation
events-generated vs invoiced → leakage assurance before close
Customer hierarchy
Platform → tenant → end-user: roll usage up the tree, rate at any level
Webhooks back
period.closed, overage.threshold, credit.low, ingest.gap
The same meter that controls your cost now bills your customers.
Cost on one side of the ledger, revenue on the other. Explore the internal dimensions, or see the neutrality thesis.
Frequently asked questions
Isn't this just Stripe's LLM Token Billing?+
No. Stripe auto-prices two or three providers and settles on Stripe's rails. We're reseller-native (bill your customers' customers up a hierarchy), all-provider (Azure, Bedrock, self-hosted, fine-tunes, embeddings, caching — not just the auto-priced big three), and processor-neutral: your meter projects to Stripe, your ERP, or a CSV, chosen per customer. You can even feed Stripe's meters from us. We're the neutral meter, not another set of rails.
We already meter tokens — why do we need you?+
Metering is the easy half. The hard half is rating it (cost → price with your margin, across every provider), billing overage, burning down prepaid credits with correct deferred-revenue treatment, and reconciling events against invoices before you close the month. That's what leaks 3–9% of revenue. You keep your meter; we turn it into accurate, reconciled, defensible billing.
Do you touch our money?+
No. We compute what each customer owes and hand you the itemized, reconciled result. Stripe or your ERP collects. We're an orchestration layer, not a payment processor — which is exactly why we can stay neutral and you keep control of the money movement.
What if we want to leave?+
Own your meter. Where the invoice lands is a projection you set per customer and can swap without re-instrumenting a single AI call. Export your full event stream any time. After 2026's consolidation — Metronome into Stripe, Orb into Adyen, m3ter into Salesforce — that neutrality is the point: if leaving requires re-instrumenting your product, you're not a customer, you're collateral.
How fast is the integration?+
Days, not a quarter. If you already write a usage row per call, we tail that table on a cursor — zero changes to your AI-call path. Otherwise it's one line after your existing metering hook. A design partner reselling across 32+ models integrated in report-mode without touching how they call LLMs.