GitHub Is Shutting Down Spark. Here's What to Export Before August 31.

What happened

GitHub is shutting down Spark, the tool it built for turning a plain-English prompt into a working app directly on github.com. As of August 4, 2026, Spark stopped accepting new users and no longer lets anyone start a new app. Existing users have until August 31, 2026 to get their code out before editing access closes for good.

Apps you’ve already deployed through Spark keep running after the shutdown. What goes away is the ability to keep editing them inside Spark’s own workbench (its browser-based build environment). To keep working on an app, open its workbench, click the ”…” menu, and choose “Create repository.” That copies your app’s code into a normal GitHub repository you can keep developing anywhere, including inside a coding agent.

There’s a second, quieter cutoff already behind us. Spark apps could call a built-in llm() function to add AI features without the builder wiring up their own API key, powered by GitHub Models, GitHub’s own AI inference service. GitHub Models retired on July 30, 2026, so any llm() call in a Spark app has already stopped working. Apps that never used it aren’t affected. If yours does, you’ll need to swap in your own AI provider’s API key before those features work again.

Why GitHub is doing this

GitHub’s own explanation is direct: AI models and agentic development tools have gotten strong enough since Spark launched that builders can now do the same idea-to-app work inside the coding surfaces they already use, specifically VS Code, the Copilot CLI (Copilot’s command-line tool), and the GitHub Copilot app. The company says it’s seeing builders choose those integrated paths over a separate, standalone builder, and it’s aligning its own product lineup around that.

Spark competed in the same space as Lovable, v0, Bolt, Replit, Firebase Studio, and Base44: type what you want, get a working app back, without opening a separate code editor. GitHub’s retreat from that model is a specific, public claim about how far agentic coding tools have come, not a verdict on that category as a whole. Whether Lovable, v0, Bolt, or other standalone builders see the same shift in how their own users work is a separate question this shutdown doesn’t answer on its own.

What isn’t clear from GitHub’s announcement is whether any Spark-specific capabilities, like its generative interface workflow, carry over once your code is a plain repository. GitHub’s guidance stops at “export it and keep going,” without detailing what changes once you’re outside Spark’s workbench.

Who should care

This is a hard deadline if you have an active Spark app: export it by August 31 or lose the ability to keep editing it. It’s a smaller but still relevant signal if you’re choosing between a standalone AI app builder and building inside a coding agent you already use for other work. GitHub, a company with visibility into how millions of developers actually build, just bet on the second approach for its own product line.

What builders should do next

If you have a Spark app, open its workbench now, use the ”…” menu to create a repository, and don’t wait for the deadline to find out whether something breaks in the export. Plan on adding your own AI provider’s API key too if the app calls llm(), since you’ll need one before those features work again.

If you’re not a Spark user but you’re weighing a standalone app builder against building inside a coding agent, run one real test before deciding. Pick a small, concrete feature, like “add a login form with email and password fields, validate the email format, and show an error message on failure,” and build it both ways. Compare three things: how long each gets you a working first version, how many follow-up prompts it takes to fix what the first attempt got wrong, and whether you can still edit the result a month from now without being locked into that tool. That last question is exactly what current Spark users are being forced to answer today.

End of article