GitHub releases a preview of Copilot SDK for Java with support for your own AI keys

GitHub released Copilot SDK for Java in version 1.0.7-preview.1, available as the Maven dependency com.github:copilot-sdk-java. According to GitHub, this is the first truly framework-agnostic way to control AI from Java – unlike Langchain4j or Spring AI, where developers had to depend on a specific library or, in the latter case, the design decisions of Spring.

The SDK supports BYOK (Bring Your Own Key) and is vendor-neutral: using ProviderConfig with your own baseUrl and apiKey (or bearer token), you can connect any model provider, including OpenAI, Anthropic, Azure or OpenAI-compatible endpoints, without needing a GitHub Copilot subscription. The library allows you to create agent sessions, register tools, send prompts and receive structured responses from server-side Java code. It works in both Jakarta EE and Spring environments and uses CompletableFuture, annotations, lambda expressions and virtual threads.

GitHub demonstrates the SDK in a sample application for managing real estate inquiries built on Jakarta EE 11 (Faces 4.1, CDI 4.1, WebSocket 2.2, Data 1.0, Persistence 3.2), running on Open Liberty 26.0.0.5 with a PrimeFaces 15.0.16 user interface and an H2 database. The application processes customer queries through an agent pipeline running on virtual threads, with real-time status updates via Jakarta WebSocket. A key SDK feature is the @CopilotTool annotation, which turns an ordinary Java method into a tool that a model can call – the SDK handles JSON Schema generation, argument parsing and argument passing. According to GitHub, this annotation API is experimental and requires enabling the -Acopilot.experimental.allowed=true flag in the Maven build.

The rest of the source article was not available.