Manage Azure AI Projects with .NET SDK

This skill covers Azure.AI.Projects, the high-level .NET SDK for Azure AI Foundry project operations (GA v1.1.0, preview v1.2.0-beta.5), covering agents, connections, datasets, deployments, evaluations, and indexes through a single AIProjectClient exposing sub-clients: Agents (versioned agents), Connections, Datasets, Deployments, Evaluations, Evaluators, Indexes, Telemetry, an OpenAI sub-client (preview), and GetPersistentAgentsClient() for low-level agent operations. The source documents eight numbered workflows in total; six of them are covered here in working detail. Persistent agents: get a PersistentAgentsClient from the project client, create an agent with a model/name/instructions, create a thread and message, run it, poll RunStatus until it's no longer Queued or InProgress, read back message content, then delete the thread and agent. Versioned agents with tools (preview): define a PromptAgentDefinition with a web-search tool scoped to a user location, create an agent version, get a ProjectResponsesClient for that version, and generate a response. Connections: list all, get one by name with or without credentials, or fetch the default connection. Deployments: list all, filter by model publisher, or get one by name. Datasets: upload a single file or an entire folder (with a regex file-pattern filter) as a named, versioned dataset, then retrieve or delete it. Indexes: create or update an AzureAISearchIndex tied to an AI Search connection and index name, then list indexes.

Use it for high-level Azure AI Foundry project operations from .NET - managing agents (persistent or versioned), connections, model deployments, training/evaluation datasets, evaluations, and search indexes within a single project. For low-level, fine-grained agent control beyond what AIProjectClient's Agents sub-client offers, use the separate Azure.AI.Agents.Persistent package directly via GetPersistentAgentsClient().

Given a PROJECT_ENDPOINT and DefaultAzureCredential, it produces an AIProjectClient whose sub-clients accept typed requests (model/instructions for an agent, a file or folder path for a dataset, an AzureAISearchIndex definition) and return typed resources (PersistentAgent, ThreadRun, AIProjectConnection, AIProjectDeployment, AIProjectDataset, AIProjectIndex).

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

Optional packages extend it: Azure.AI.Projects.OpenAI for versioned agents with OpenAI-style tools like web search, and Azure.AI.Agents.Persistent for low-level agent operations. Authenticates via DefaultAzureCredential against PROJECT_ENDPOINT, with MODEL_DEPLOYMENT_NAME, CONNECTION_NAME, and AI_SEARCH_CONNECTION_NAME as supporting environment variables.

.NET developers building on Azure AI Foundry who need one client for agent orchestration, connection management, model deployment discovery, dataset upload/versioning, and Azure AI Search index creation within a project, rather than wiring up each Azure AI service's SDK separately. Evaluations and Evaluators sub-clients, plus a dedicated Azure OpenAI chat client accessor, round out the surface beyond the six workflows detailed above.