Launch a tenant offering and invoice it

Resources, then Rate Cards. Every model you serve or resell carries a rate per million tokens in and out. These rates are what a customer's calls are priced at, whichever organisation makes them.

Resources, then Quotas. A template is the compute envelope a customer organisation is born with: CPU, memory, storage, pods, and what that envelope costs you a month. Small team is the one this offering uses.

Tenant Management, then Organizations. Five organisations already run on this platform. Click New tenant.

The customer's name, as it will read on their platform and on their invoice. We typed one; change it if you like.

The slug names the organisation everywhere: its namespace, its sign-in realm, its rows in billing. Checked live for availability.

The person you are inviting. They get a one-time password at the end of this wizard and run their own organisation from then on.

The subdomain the organisation signs in at, picked from the domains you registered in DNS Management. We picked the spare one.

Click Next.

The template from the Quotas page. The organisation's namespace is provisioned with this envelope, and billing prices it from the quota ledger.

Models as a service: the organisation calls these on your provider accounts and your deployments, at your rate card, and you bill them. We searched the catalogue for the first one.

Click qwen3-32b.

A larger model for the harder questions. Same gateway, same rate card, same bill.

Click llama-3.3-70b. Anything not on this list is refused by the gateway, with the reason.

Click Next.

Company, slug, admin, quota, the entitled models, the production environment. This is what you are selling. Click Create organization.

The organisation exists: realm, namespace, production environment. The admin's temporary password shows once and is never stored. Tick I have saved this password.

Click Done.

One user, no GPUs held, the production environment provisioning. Click Policy.

The two models from the wizard, held as this organisation's own bundle. Add one here later and the gateway honours it on the next call; remove one and it is refused.

The offering carries a monthly spend cap, warning at eighty percent. The organisation's first calls are already counted against it; on this demo they are simulated the moment the entitlement lands.

You are the customer's admin, signed in to their own platform under the provider's name. AI Gateway lists exactly the models the offering entitled, at the rates you set, with the provider named as the one who entitled them.

Click the workspace switcher.

Work Room, Control Room, Catalog. No Studio: building agents is the next offering, agents as a service, and it is off for this organisation until the provider switches it on. Absent, not advertised.

Back as the provider. Billing & Usage lists every tenant with what it consumed this period at the rates in force; the new organisation's first calls are already on the row.

Click Generate invoices.

A run can cover the whole estate or one customer. The period is the last closed month; a month still accruing is refused. We picked the new organisation.

Click Generate. A tenant with no usage signal is never invoiced at zero, and a tenant already invoiced for the period is skipped.

Click Invoices.

A numbered invoice for the customer, for the period, priced from metered usage against your rate card: metered spend, discount, total, due date, status. Issued, never born paid.

Operations, then Usage Analytics: seats in use, recorded cost from billing, and adoption counted from what the offering entitles, per tenant.

Tenant Management, then License & Entitlements. The new organisation took one of your licensed tenants. The apps pool below is the next offering, agents as a service: licensed to you already, sold to a customer when you switch Studio on for them.