providers/azure · Azure AI Project

What this lesson demonstrates

This lesson wires an agent to an Azure AI Foundry project — not a bare model deployment, but a Foundry project endpoint addressed through the project Responses API. You give foundryprovider.NewAgent three things: the project endpoint, a TokenCredential, and a ModelDeployment target naming the deployed model. The provider derives the project’s OpenAI-compatible base URL (…/openai/v1/) from the endpoint and talks to it. From there the agent is identical to every other lesson.

The distinguishing detail is the mode selector. foundryprovider.ModelDeployment(model) tells the provider to treat the endpoint as a project endpoint and use project Responses API mode, rather than a plain chat-completions endpoint.

One real excerpt

Construction is factored out of main so the offline test can build the same agent with a fake credential:

func newJoker(endpoint, model string, cred azcore.TokenCredential, mw agent.Middleware) *agent.Agent {
    return foundryprovider.NewAgent(
        endpoint,
        cred,
        foundryprovider.ModelDeployment(model), // project Responses API mode
        foundryprovider.AgentConfig{
            Instructions: "You are good at telling jokes.",
            Config: agent.Config{
                Name:        "Joker",
                Middlewares: []agent.Middleware{mw},
            },
        },
    )
}

What to notice — the gotcha

ModelDeployment(model) is the mode switch, and it changes how the endpoint is interpreted. Pass it and the endpoint is treated as a Foundry project endpoint, from which the provider builds the OpenAI-compatible base URL itself. Get the target wrong and you’ll point a project-mode agent at a non-project URL (or vice versa) and requests will fail in confusing ways. The endpoint, model, and credential all come from the repo’s own internal/demo helper (Endpoint(), Model(), Credential()), which follows the standard Foundry env-var convention — FOUNDRY_PROJECT_ENDPOINT and FOUNDRY_MODEL.

The credential is lazy: demo.Credential() builds a TokenCredential that doesn’t call Azure until the first request, so the structural test builds the whole agent with a fake credential and a dummy endpoint and asserts its wiring with no network.

How it maps to the Agent Framework

This is the canonical “run against my Foundry project” path. In Azure AI Foundry, a project is the unit that groups deployed models, connections, and settings; addressing the project endpoint (rather than a raw model URL) is what lets Foundry apply project-level configuration. The provider hides the base-URL derivation and authentication scope entirely — you supply the project endpoint and a deployment name, and the same RunText/Collect surface works unchanged. The ModelDeployment target is the one knob that distinguishes this from the other Azure provider lessons in the series.

Run it

go run ./tutorial/02-agents/providers/azure/ai_project

Live runs need az login plus FOUNDRY_PROJECT_ENDPOINT (and FOUNDRY_MODEL). The structural test runs offline with a fake credential; live calls are gated behind AF_LIVE=1.

Next: Azure · Foundry Model