Azure AI Foundry projects

Discover agents that are deployed in your Foundry project and monitor these deployments in the registry UI.

Before you begin

Set up an OIDC provider. This guide assumes that you installed Keycloak. Make sure to follow the General setup tab to install Keycloak with the required realm settings, clients, and secrets.
Install Solo Enterprise for agentregistry.
Connect Solo Enterprise for agentregistry to your Azure AI Foundry project.
Note
Solo Enterprise for agentregistry does not create or deploy AI Foundry agents. The MicrosoftFoundry runtime is a discovery-only runtime.

Step 1: Create a Foundry agent

Create an agent and deploy it to your Foundry project. The following steps show how to create agents by using the Azure portal or a Python script.

Open the AI Foundry home page ai.azure.com and select your Foundry project from the project list.
Open the Models view, select a model, such as gpt-5.4, and click Deploy.
Open the Agents view and click New agent → Build an agent.
Enter a name for your agent and click Create.
Configure your agent.
1. In the agent overview page, select the model deployment that you want your agent to use.
2. Add instructions for your agent. For example, you can use the following instruction.
  You are a concise test agent for AgentRegistry discovery.
3. Save or create the agent.
Open the agent Playground and send a test message to confirm the agent runs. For example, you can ask the agent what it can do for you.

Note

To create agents with the Python SDK, your personal az login identity must have the Azure AI Developer role on the Foundry project. This permission is separate from the Foundry User role that you granted to the Entra app registration earlier as this role is used by Solo Enterprise for agentregistry to discover deployed agents.

Use this approach for a repeatable, scripted setup. Requires azure-ai-projects>=2.0.0.

Set up a Python virtual environment and install the required packages.

python -m venv .venv
source .venv/bin/activate
pip install azure-ai-projects azure-identity

Log in to Azure.
```
az login
```
Set an environment variable for the model deployment name. You can find available model deployment names under Models → Deployments in your Foundry project.
```
export MODEL_DEPLOYMENT_NAME=<model-deployment-name>
```

Create a Python script file. The file completes the following tasks:

Connects to your Foundry project by using your az login credentials.
Creates a minimal prompt-based agent that is named agentregistry-smoke-test and backed by your deployed model.
Prints the agent name and version on success.

cat > create-foundry-agent.py << 'EOF'
import os
from azure.ai.projects import AIProjectClient
from azure.ai.projects.models import PromptAgentDefinition
from azure.identity import DefaultAzureCredential

with AIProjectClient(
    endpoint=os.environ["AZURE_PROJECT_ENDPOINT"],
    credential=DefaultAzureCredential(),
) as project:
    agent = project.agents.create_version(
        agent_name="agentregistry-smoke-test",
        definition=PromptAgentDefinition(
            model=os.environ["MODEL_DEPLOYMENT_NAME"],
            instructions="You are a concise test agent for AgentRegistry discovery.",
        ),
    )
    print(f"Created Foundry agent: {agent.agent_name} (version: {agent.version})")
EOF

Run the script.

python create-foundry-agent.py

Example output:

Created Foundry agent: agentregistry-smoke-test (version: 1)

Step 2: Verify discovery

After the runtime is created, Solo Enterprise for agentregistry triggers an initial sync to list deployed agents as Instances in the registry UI.

Note

It can take a few minutes for the sync to complete and discovered agents to show up in the registry UI.

Open Instances view in the registry UI.
From the Managed drop down, select Unmanaged, Discovered Resources.
Verify that you see the Azure Foundry agent that you created earlier.

Note

If agents do not appear, check the following in your runtime setup.

The runtime uses the Foundry project endpoint, not the account endpoint.
The Entra app has the Foundry User role assigned on the Foundry project resource, not only at the resource group or subscription level.
The secret exists and status.dataKeys includes clientSecret.
The Foundry project contains at least one agent.

For detailed error information, check the agentregistry server logs for foundry: API returned or resolve client secret messages.

Unmanaged agents from deployment runtimes Microsoft Copilot Studio