Gateway
For platform teams
€99per month
Includes 100,000 routed requests a month.
100,000 requests is about
- Per day
- 3,300
- Dispatch-200 runs
- 500
Inference is billed separately by your model providers.
Pricing
Pay for the route, not the model
One gateway fee covers 100,000 routed requests a month. Inference stays on your providers' bills, at their rates.
- Every policy, route and report is in the fee
- No markup on provider charges
- A word from us before any change to your bill
One month, two ledgers
A sample month at the Dispatch-200 mix. Joulis bills the gateway; your providers bill each model call.
Sample month14 Oct to 13 Nov
100,000requests through the gateway
- Gannet-S74,880
- Gannet-L24,960
- Exceptions before execution160
Billed by Joulis
| Line | Qty | Amount |
|---|---|---|
| Gateway fee100,000 requests included | 1 | €99.00 |
| Routed requestsGannet-S and Gannet-L | 99,840 | included |
| Exceptions before executioncounted once, never run | 160 | included |
| Policies, routes and reportsevery one you set up | all | included |
Due to Joulis€99.00
est.In the report: 24.96 kWh accelerator energy, facility overhead left out.
Billed by your providers
| Line | Qty | Amount |
|---|---|---|
| Gannet-S callsrouted by your policy | 74,880 | their rate |
| Gannet-L callsrouted by your policy | 24,960 | their rate |
| Exceptions160 never sent | 0 | €0.00 |
| Markup from Joulisnone on any call | 0% | €0.00 |
Due to your providersTheir rates × 99,840 calls
You keep your provider contracts and keys. Joulis does not resell inference.
Everything the gateway does is in the fee
Every policy, route and report comes with the €99. Nothing below is an add-on.
-
Policies with three budgets
A quality floor, a latency ceiling and an energy ceiling on every policy you write.
≥ 89/100 · ≤ 1.8 s p95 · ≤ 60 Wh est. -
Named fallback routes
One route your team chose in advance, used only when nothing else fits the policy.
fallback: gannet-l -
Exceptions before execution
When no route qualifies, the request comes back unrun, with the budget it would break.
exception: no_route_meets_policy -
Result metadata on every response
The route taken, its p95 latency and its estimated energy, beside each result.
route · p95 · Wh est. -
Benchmarks on your workload
Two policies on identical tasks and limits, scored on a fixed rubric, as in Dispatch-200.
same tasks · same limits -
Monthly ledger exports
Every request, route and estimate of the month, as CSV or JSON, ready for the review.
csv · json
Not in the fee: inference. Your model providers bill each call at their rates, on their own invoices.
How energy is estimatedBilling, line by line
When the month starts, what counts toward 100,000, and what never reaches the Joulis bill.
01When does the billing month start?
On the day your first policy routes a live request. If that was 14 October, each month runs from the 14th to the 13th, and the 100,000 requests reset on the 14th.
02What counts as a request?
Every call that reaches the gateway with a policy id counts once, whichever route serves it. A retry you send is a new request. A fallback Joulis takes inside the same request is not.
03Does an exception before execution count?
Yes, once. The gateway evaluated your policy and returned the exception, so it counts toward the 100,000. It never reaches a provider, so no inference is billed for it.
04What happens past 100,000 requests?
Requests keep routing and the ledger keeps counting. Nothing changes on your bill without a word from us: a person from Joulis contacts you before any change to your plan or fee.
05Do you mark up provider charges?
No. Joulis does not resell inference. Your providers bill you directly at their rates, under your own contracts, and the Joulis bill holds one line: the gateway fee.
Another billing question? access@joulis.io
Put a budget on your next rollout
Bring one workload. We will run it under your policy and show you what each route delivers and consumes.