The AI lab

To try an AI model, there is nothing to create and nothing to request: an Azure AI Foundry project is already there. You open the portal with your Qim info account, chat with a model in the playground, and create your own project the day you want to deploy a model and call it from your code.

Open the labs Foundry project →Playground, model catalog and deployments, with your Qim info account.

What is already in place

What exists	What it means for you
The ai.azure.com portal	The single entry point: model catalog, playground, deployments, agents. You sign in with your Qim info account, so there is no key to obtain before you start.
A Foundry resource aif-azurefoundry-lab-szn, in Switzerland North	It holds the models, the quotas and the keys. It lives in the Resource Group rg-azurefoundry-lab-szn, created and maintained by the administrators: no RG to create, no tags to set, no declaration e-mail.
A shared project proj-azurefoundry-lab-szn	Everything you need to try a model right away, without creating anything. It is a common project: what you deploy there is visible to your colleagues.
The right to create your own projects	As soon as the test goes beyond a couple of prompts in the playground: your project, your deployments, your endpoint. See create my project.
The model catalog and deployments	You pick the model and deploy it yourself. The catalog shows what is available in the region, and what each model costs per token.

This is the right lab if

The question	The answer that brings you here
What is the test about?	A model: a system prompt, a comparison between two models, an agent, an integration called from your code.
What needs hosting?	Nothing. A model deployment, and your code stays on your machine or in another lab.
Do you need an Azure lab?	No. The resource exists and is shared: no Resource Group, VM or network to set up.
Is it free?	No, and that is the difference with the Kubernetes lab: every model call is billed per token.

What you type leaves your machine

Prompts and responses are processed by the Azure service. They are not used to train the models and are not shared with model providers, but a sample may be retained and reviewed by Microsoft for abuse detection. A Global deployment may also process your prompts outside Switzerland. The “no client, personal or sensitive data” rule therefore applies to the content of your prompts, not just to files.

Here, usage is what costs

No VM to shut down, so no 7 p.m. auto-shutdown to protect you: tokens are the meter. An unused standard deployment costs nothing, but a provisioned deployment is billed by the hour even if no call is made, and a runaway loop burns through tokens fast. See Monitoring costs and Microsoft Foundry pricing.

The rules don’t change

A shared, ready-to-use resource is still a lab.

Applicable rules

No production use

My responsibility

No production, critical or long-lived use is allowed in a lab.

MandatoryPR-GOVCLOUDLAB-A-01 / PR-GOVLABONPREM-A-01

No sensitive data

My responsibility

Never put real client data, personal data or any sensitive data in a lab.

MandatoryPR-GOVCLOUDLAB-A-01 / PR-GOVLABONPREM-A-01

Indicative budget

Recommendation

Indicative budget of about 200 CHF per month per lab.

Not blockingPR-GOVCLOUDLAB-A-01

Clean up and document

My responsibility

Stop or delete unused resources, and document the lab as you go.

Not blockingPR-GOVCLOUDLAB-A-01

Where to start

Try a model

The shared project’s playground: a prompt, a response, a comparison. Allow five minutes, with nothing to install.

Try a model →

Create my project and call the API

Your project, your model deployment, the endpoint, and a call from your code.

Create my project →

A need beyond this scope — a model missing from the catalog, a larger quota, a provisioned deployment, fine-tuning: write to lab@qiminfo.ch describing the test and how long it will last.