AI 開發工具
GitHub Spark Stops Creating New Apps; Existing Projects Using llm() Must Replace Their Inference Service
GitHub has stopped accepting new Spark users and allowing new apps to be created. Existing users have until August 31 to export their code. Deployed sites can continue operating, but llm() calls that depend on GitHub Models have stopped working, requiring developers to integrate an external model provider and take over API key and billing management.
GitHub began winding down Spark on August 4: the platform no longer accepts new users or allows new apps to be created. Existing users can access the workspace until August 31; after that, they must first turn their project into a GitHub repository to continue making changes. GitHub says deployed Spark apps will continue to operate after the editing interface is retired, so not every app will go offline immediately. However, development and runtime dependencies must be evaluated separately.
The most significant breaking change has already occurred. Spark previously provided direct access to GitHub Models through `llm()`/`spark.llm`, eliminating the need to select a provider, configure an API key, or manage separate billing. GitHub Models was retired on July 30, so those calls no longer work. GitHub recommends searching the codebase for `llm()` before exporting: apps without this call are unaffected by the inference service retirement, while those that use it must switch to their own model provider and handle secret storage, usage limits, error retries, and billing themselves.
The change also exposes portability issues with AI app generators. According to the original [product documentation](https://docs.github.com/en/copilot/concepts/spark), Spark uses React, TypeScript, GitHub authentication, Azure Container Apps, and managed key-value storage backed by Azure Cosmos DB. Exporting the source code does not mean all these managed capabilities will automatically be converted into equivalent components deployable in another environment. Teams should inventory SDK calls, the data persistence layer, authentication flows, and deployment configuration—not merely verify that the frontend compiles.
GitHub is directing future development toward VS Code, Copilot CLI, and Copilot app, but the [deprecation announcement](https://github.blog/changelog/2026-08-04-upcoming-deprecation-of-github-spark-on-github-com/) does not promise one-click migration or a compatibility layer. Engineering teams should first export their projects and establish reproducible build processes, then add regression tests covering model outputs, permissions, and data migration. In particular, they should not interpret “deployed apps will continue to operate” as a long-term guarantee for their AI functionality or backend dependencies.