How it works · AI Universe

One deployment per organisation, entirely inside your own Azure subscription. Here is what it is made of and how a question reaches a model.

AI Universe is installed once per organisation, in a subscription you own. It is not a shared cloud service: every part of it, including the models, runs in your tenant.

The parts

Part	What it does
Control panel	Where admins and app owners create apps, choose models, manage people and follow cost. Sign-in is with your own Entra ID.
Apps	The AI apps your teams use. Each app has its own code repository, its own cloud resources, its own people and its own agent. In the platform's technical pages an app is also called a galaxy.
Kernel	The engine behind the control panel. It creates apps, runs upgrades and talks to Entra ID and Azure on the platform's behalf.
AI gateway	Every call to a model goes through it. It holds the only credential for the models, applies the safety rules and records usage for cost reporting.
Microsoft Foundry	Hosts the models and agents, in your subscription.

flowchart LR
    people(["Your people"]) --> apps["Apps"]
    admins(["Admins"]) --> panel["Control panel"]
    panel --> kernel["Kernel"]
    apps --> gateway["AI gateway<br/>safety rules, usage"]
    gateway --> foundry["Microsoft Foundry<br/>models and agents"]
    kernel --> entra["Entra ID"]
    kernel --> azure["Azure"]
    subgraph tenant["Your Azure tenant"]
        apps
        panel
        kernel
        gateway
        foundry
        entra
        azure
    end

What reaches Fortytwo

Only two things cross the boundary of your tenant:

Releases. New versions are copied from Fortytwo's release registry into your deployment's own registry when you upgrade.
The licence check.

Your data, prompts, replies and logs stay in your subscription. Fortytwo cannot sign in to your deployment.

How an app reaches its users

Each app gets its own address under a domain you choose at install, such as hr-assistant.ai.yourcompany.com. All traffic enters through one Application Gateway with a web application firewall. Nothing else in the deployment is reachable from the internet.

People sign in with their normal work account. An app only lets in the people and groups you give access to on that app's People page.

How a question reaches a model

Someone asks an app's agent a question.
The app sends it to the AI gateway, never straight to a model. Apps hold no model keys.
The gateway checks the question against your safety rules, records usage, and passes it to the model.
The reply comes back through the same rules before the person sees it.
If the agent needs to look something up, it calls one of the app's own tools. The platform only lets it use the tools you allowed on the Agents page.