Manage Azure AI Projects with .NET SDK

This skill teaches an agent to use the Azure.AI.Projects .NET SDK, the high-level client for Azure AI Foundry project operations: agents, connections, datasets, deployments, evaluations, and indexes. AIProjectClient is the entry point, exposing sub-clients for Agents, Connections, Datasets, Deployments, Evaluations, Evaluators, Indexes, Telemetry, a preview OpenAI client, and GetPersistentAgentsClient() for low-level agent operations - create an agent, create a thread, post a message, create and poll a run, read messages, then delete the thread and agent. A preview path (Azure.AI.Projects.OpenAI) builds versioned agents with tools such as a web search tool, and exposes a ProjectResponsesClient for calling them.

Use it when writing .NET code against an Azure AI Foundry project that needs agents, model deployments, datasets, search indexes, or evaluations through one client rather than calling each Azure AI service's own SDK separately. Reach for the low-level PersistentAgentsClient (via GetPersistentAgentsClient()) for simple agent, thread, and run workflows, and the preview Azure.AI.Projects.OpenAI versioned-agent path for production scenarios that need tools. It needs an Azure AI Foundry project endpoint and Azure credentials (DefaultAzureCredential recommended for production); it is not the right skill for the Python or JavaScript SDKs, which ship as separate packages.

Inputs: a PROJECT_ENDPOINT, a MODEL_DEPLOYMENT_NAME, and connection names (CONNECTION_NAME, AI_SEARCH_CONNECTION_NAME) as environment variables, plus a DefaultAzureCredential. Depending on the operation: agent name, instructions, and model plus thread messages for agent runs; dataset name, version, and file or folder path for uploads; index name, version, and an AzureAISearchIndex for search indexes; an Evaluation with an InputDataset and EvaluatorConfiguration map for evaluations. Outputs: typed results per sub-client - PersistentAgent, ThreadRun, and message content for agents; AIProjectConnection, AIProjectDeployment, AIProjectDataset, and AIProjectIndex objects for listing and metadata calls; and Evaluation objects carrying status and results, with listing operations exposed as AsyncPageable<T>.

Azure AI Foundry (agents, connections, deployments, datasets, indexes, evaluations), Azure Identity (DefaultAzureCredential), Azure AI Search (AzureAISearchToolDefinition / AzureAISearchIndex), Bing Grounding (BingGroundingToolDefinition), Azure Functions (AzureFunctionToolDefinition), MCP (MCPToolDefinition), OpenAPI-defined external APIs (OpenApiToolDefinition), and Azure OpenAI chat completions through AzureOpenAIClient.

dotnet add package Azure.AI.Projects
dotnet add package Azure.Identity

# Optional: For versioned agents with OpenAI extensions
dotnet add package Azure.AI.Projects.OpenAI --prerelease

# Optional: For low-level agent operations
dotnet add package Azure.AI.Agents.Persistent --prerelease

.NET developers building against Azure AI Foundry who need one client for agents, connections, deployments, datasets, indexes, and evaluations, from quick agent and thread prototypes up to versioned, tool-using agents in production.