GitHub Is Shutting Down Spark on August 31

GitHub Spark is done. Starting August 4, 2026, new users can’t sign up and no new apps can be created. Existing users have until August 31 to export anything they want to keep.

This follows GitHub Models retiring on July 30. That already broke Spark apps that called the llm() function, since Models was the inference layer powering those calls. If your Spark app uses llm(), it stopped working last week. Apps that don’t touch llm() still run, but they lose access entirely at month’s end.

What was Spark

GitHub launched Spark as a way to go from a natural-language description to a running micro-app without writing code. It combined a visual editor, a hosted runtime, and GitHub Models as the inference backend. The pitch was simple: describe what you want, get a working app.

It lived somewhere between v0, Bolt, and Firebase Studio, but inside GitHub’s ecosystem instead of a standalone product.

Why it’s being shut down

GitHub’s stated reason is that the underlying tools have outgrown what Spark was built around. “AI models and agentic development tools have advanced significantly,” the announcement says. “Builders can now build and refine these experiences through GitHub Copilot in the environments where they already work.”

That’s not wrong. Copilot’s cloud agent has gotten significantly more capable since Spark launched, and the Copilot app (released in June) gives users a dedicated interface for running agents against repositories. GitHub’s argument is essentially that Copilot does what Spark was doing, but better and in places developers already are.

What to do before August 31

If you have Spark apps you care about, export them now. Open the Spark workbench for each app, hit the ... menu, and choose “Create repository.” That puts the code in your GitHub account.

Note that exporting gives you the code, not a working hosted app. If your app used llm(), you’ll need to wire it up to an inference provider yourself — GitHub Models is gone, so there’s no drop-in replacement in the GitHub ecosystem. OpenAI, Anthropic, or Google’s APIs are the obvious options.

Apps without AI features are simpler to migrate. The code runs fine; you just need somewhere to host it.

Copilot gets new automation controls the same week

The shutdown news landed alongside a batch of Copilot improvements that suggest where GitHub’s agentic focus is going instead.

As of August 3, you can trigger Copilot cloud agent automations from issue and PR comments. You configure what comment text triggers which automation, and the agent spins up in its own cloud sandbox when it sees a matching comment. The obvious use cases: commenting on an issue to kick off an investigation of a stack trace, or commenting on a PR to auto-generate updated documentation based on the diff.

Also shipping this week: adjustable reasoning levels for the Copilot cloud agent when delegating tasks. For models that support reasoning control, you can now tell the agent to think harder or lighter depending on what the task warrants. And enterprise admins can target specific teams with customized settings, rather than applying one config across the whole org.

None of these are dramatic changes, but they fill in gaps that made automation workflows more rigid than they needed to be.

The pattern

GitHub has now retired GitHub Models, GitHub Spark, and GitHub’s standalone model playground in a span of a few weeks. What’s left standing is Copilot: the VS Code extension, the web interface, the CLI, the cloud agent, and the dedicated desktop app.

This looks less like trimming fat and more like a deliberate consolidation. GitHub built a lot of AI surface area quickly, and now it’s funneling everything into the product that has enterprise paying customers and real adoption. Spark was a useful experiment. It didn’t become the main thing.

Copilot did.