foundry-sdk

Foundry SDK

WHEN: Writing implementation code against Microsoft Foundry SDKs or Azure AI Projects clients for agents, evaluations, datasets, indexes, connections, tracing, or SDK-managed tools.

When to Use

• Creating agents programmatically from a Foundry project client
• Wiring Foundry tools such as file search, MCP, Azure AI Search, or Azure Functions
• Running evaluation jobs, datasets, and indexes from code
• Listing or validating model deployments and project connections
• Enabling SDK-level tracing and Azure Monitor observability

Decision Tree

Working with Microsoft Foundry?
+- Architecture and model strategy only?
|  - Use azure-foundry
+- Need implementation against SDK clients and APIs?
|  - Use foundry-sdk
+- Building agents with Agent Framework abstraction?
|  - Combine foundry-sdk with ai-agent-development
+- Need evaluation jobs, datasets, indexes, or connections?
|  - Use foundry-sdk with ai-evaluation
-- Need deployment/operational portal workflows?
   - Use Azure MCP operational guidance alongside this skill

Core Rules

1. Use Entra ID and project endpoints; do not hardcode secrets or long-lived keys when SDK auth supports credentials.
2. Treat deployments, datasets, indexes, and connections as versioned resources with explicit ownership.
3. Keep agent definitions, prompts, and tool schemas in repo files; do not bury them inside client-construction code.
4. Validate deployed model names and project connections before runtime traffic depends on them.
5. Enable tracing intentionally and review privacy implications before propagating trace context or content.

Python Baseline

Microsoft docs currently position azure-ai-projects as the Python client library for Microsoft Foundry project operations.

import os
from azure.ai.projects import AIProjectClient
from azure.identity import DefaultAzureCredential


with (
    DefaultAzureCredential() as credential,
    AIProjectClient(
        endpoint=os.environ["AZURE_AI_PROJECT_ENDPOINT"],
        credential=credential,
    ) as project_client,
):
    for deployment in project_client.deployments.list():
        print(deployment.name)