Multi-Tenant Model Access, Metered and Billable
You already run a gateway in front of your models. What it does not give you is a way to sell what sits behind it: per-tenant isolation, per-tenant keys, quotas that actually hold, spend attributed to the right customer, and an invoice at the end of the month.
The Hybr® LLM Gateway registers the gateway deployment you already operate and turns it into a multi-tenant service. Each tenant gets an inference endpoint that works with the standard OpenAI SDK — for most customers a base URL and an API key change, nothing more.
What you get
- A tenant endpoint per customer. OpenAI SDK-compatible, so existing application code works against it.
- Per-tenant keys and model entitlements. Decide which models each tenant can reach, and revoke access without touching the gateway.
- Quotas and budget ceilings. Stop one tenant exhausting shared capacity, and stop a runaway agent turning into an unbilled surprise.
- Spend per tenant and per model. Total spend is a first-class column, not something reconstructed from logs.
- Health and model inventory for every registered gateway instance, per subscription.
- Rated usage that lands on the invoice alongside Microsoft CSP, cloud consumption and everything else you sell.
How registration works
Registering a gateway takes three pieces of information:
- A connection name.
- The admin API endpoint and master key, so Hybr® can manage tenants and keys on your gateway.
- The tenant endpoint — the URL your customers will use with the OpenAI SDK for inference.
Connectivity is either direct or via Hybr Edge. Edge mode is for gateways on a private network that should not be publicly reachable — the gateway stays where it is and Hybr® reaches it through the edge connection.
What Hybr® does not do
Worth being explicit, because it is the first question every technical buyer asks. Hybr® does not host your models and does not replace your gateway. Inference runs where it runs today, on your hardware or your cloud account, under your data controls. Hybr® sits in front of it and adds the multi-tenancy, governance, metering and commercial layer that turns it into a service.
That boundary is deliberate. It means you can change model providers, move inference between sites, or renegotiate your GPU supply without any of it reaching your customers’ integrations or your billing.
Who this is for
- Service providers reselling inference to multiple customers who need isolation and per-customer billing.
- Enterprises running a central AI platform and charging it back to business units.
- Sovereign and regulated operators who must keep inference in-country but still need commercial multi-tenancy.
Frequently asked questions
Do our customers have to change their code?
In most cases it is a base URL and an API key change. The tenant endpoint is used with the standard OpenAI SDK.
What if our gateway is on a private network?
Register it in Via Hybr Edge connectivity mode rather than direct. It does not need to be publicly reachable.
Can we run more than one gateway?
Yes. Instances are registered per subscription, so you can separate by region, by environment, or by customer segment and still bill through one engine.
Can AI usage appear on the same invoice as Microsoft CSP?
Yes. That is the point of running it on the same commerce engine — one consolidated invoice per customer across every service you provide.
Related: AI Factory · AI Application & Token Billing · Kubernetes & GPU-as-a-Service · CSP Billing Ultimate
See it against your own gateway
The fastest way to judge this is with your own gateway and your own tenant model. Book a working session and we will register one and walk through it.