Manage Azure AI Projects with Java SDK

Azure AI Projects SDK for Java is a high-level SDK for Azure AI Foundry project management, giving access to connections, datasets, indexes, and evaluations, authenticated via AIProjectClientBuilder with DefaultAzureCredentialBuilder against a project endpoint shaped like https://<resource>.services.ai.azure.com/api/projects/<project>.

A single builder produces seven sub-clients: ConnectionsClient (enumerate connected Azure resources), DatasetsClient (upload documents and manage datasets), DeploymentsClient (enumerate the AI model deployments available to the project), IndexesClient (create and manage search indexes), EvaluationsClient and EvaluatorsClient (run and configure AI model evaluations), and SchedulesClient (manage scheduled operations). Connections are listed with their name, type, and credential type, letting you check the credential type before using it. Indexes can be listed by their latest version (name, version, description) or created and updated by pointing an AzureAISearchIndex at a named search connection and search index, versioned explicitly (for example, version "1.0") rather than overwritten in place. Evaluations are handled by exposing OpenAI's own official SDK (EvalService) directly through evaluationsClient.getOpenAIClient(), rather than reimplementing evaluation APIs from scratch.

Use it when you need a high-level Java client for Azure AI Foundry project resources - connections, datasets, deployments, indexes, and evaluations - rather than provisioning them individually through separate SDKs or the portal. It's specifically the project-management layer; running an OpenAI evaluation still goes through OpenAI's own SDK surface once you have the EvalService handle from evaluationsClient.

Input is the Azure AI Foundry project endpoint plus resource identifiers (connection names, index names/versions, dataset content). Output is enumerated or created project resources - connections, indexes, deployments, evaluations - returned as paginated results via PagedIterable. Install it with:

<dependency>
    <groupId>com.azure</groupId>
    <artifactId>azure-ai-projects</artifactId>
    <version>1.0.0-beta.1</version>
</dependency>

Its own best practices: use DefaultAzureCredential for production auth, reuse a single client builder to create multiple sub-clients efficiently, always handle pagination with PagedIterable when listing resources, use environment variables for connection names and configuration, and check a connection's type before accessing its credentials. Errors surface as ResourceNotFoundException for a missing resource or the general HttpResponseException with an HTTP status code.

It wraps Azure AI Foundry's connections, datasets, deployments, and Azure AI Search-backed indexes, and exposes OpenAI's official evaluation SDK (EvalService) directly for running model evaluations.

Java developers managing Azure AI Foundry projects - connections, datasets, model deployments, search indexes, and evaluations - who want one high-level client rather than assembling access to each resource type separately.