GitHub Spark in public preview for Copilot Pro+ subscribers

[tuves]

Stuck between idea and implementation? Spending weeks on mock ups or docs that never ship? GitHub Spark takes you from idea to deployed app in minutes.

Build and ship full-stack intelligent apps using natural language with access to the full power of the GitHub platform—no setup, no configuration, and no headaches.

Key features

• Natural language to app: Describe your idea and watch Spark build it, with frontend and backend capabilities included, all powered by Claude Sonnet 4.
• No setup required: Data, LLM inference, hosting, deployments, and GitHub auth all included out-of-the-box.
• Add AI to your apps: Add intelligent features powered by LLMs from OpenAI, Meta, DeepSeek, xAI and more - no API key management needed
• One-click deployments: Publish your app with a single click.
• Build your way: Use natural language, visual editing controls, or code with GitHub Copilot code completions at your fingertips to iterate on your ideas.
• Create a repository: Get a repository with GitHub Actions and Dependabot incorporated in just a click. Everything stays synchronized so you aren't trapped in a sandbox.
• Expand with Copilot agents: Open a codespace directly from Spark to iterate with Copilot agent mode or assign an issue to Copilot coding agent.

Get started

Spark is available in public preview for Copilot Pro+ users, with rollout to additional customers coming soon.

Visit github.com/spark to build your first app.

Copilot Pro+ subscribers receive access as part of their plan. Spark messages use premium requests included in GitHub Copilot plans.

Learn more

• View more details on the Spark product page.
• Read the Spark documentation.

Disclaimer: The UI for features in public preview is subject to change.

[CurationsLA]

Looking forward to testing this out! Just started using GitHub. Great product.

[josekurian]

Great project

[askpt]

I was part of the technical preview but I cannot access the github.com/spark only the old link.

[kelseyrae]

Granted you access! Enjoy! We're rolling out to technical preview users as quickly as possible (unfortunately we're limited by model capacity so we have to do a gradual rollout).

[McCrossin]

I have seen the email about this as well but the links don't work for me even though I'm a pro+ user.

[kelseyrae]

Can you share what you're seeing? If you are trying to open the links on mobile and you have the GitHub mobile app on your phone you may need to manually open the link in a mobile browser window.

[McCrossin]

Thanks that's probably the issue just tried checking it out on my commute.

[mohamedcherifmo]

I just registered to Pro plus and I'm unable to access the repo

[kelseyrae]

Can you share more about what errors you're seeing? You should be able to access it at github.com/spark if you have a Pro+ account.

[guygregory]

Are there any plans to (eventually) make GitHub Spark available to students?

[kelseyrae]

Yes, no concrete date yet, but we are planning to roll out to additional customer segments in the coming weeks and months.

[CurationsLA]

Remember me! Game changer.

[luyandadhlamini2]

The concept looks great. I've tested it on a sample app. Agent was able to build a usable MVP, but there are teething issues.

Current issues:

1. App keeps refreshing & can't be interacted with.
2. Clicking on publish shows a loading icon next to the word "publishing" & doesn't publish the app.
3. Attempting to open the app in a codespace failed, returning a NETWORK_ERR.
4. Since the app keeps refreshing, feedback can't be provided to the agent to make changes, I assume because the agent waits for app to load before being able to accept input.

My recommendation:
While in preview phase, assign each Spark a UUID (instead of a 20 character name as is currently done) so that we can provide our spark links for the github team to debug & see issues.

@tuves
@kelseyrae
@ghostinhershell

[mpommrehn]

What is the privacy policy for Spark? Asking relative to a corporate account that includes Copilot (not sure which level), with strict limits on AI usage, only as managed through Copilot under the MSFT Corporate/Enterprise Privacy Policy.
Need to see whether Spark prompts, code, and any generated artifacts are fully contained within the same corporate privacy data wall as the rest of our Copilot usage.
If not now, is that a future intent?
Thanks!

[between-sky-sea]

Does spark allow for custom domain?

[mgyarmathy]

Do you plan to incorporate GitHub Organizations into the Visibility settings for deployed apps? Anyone with the link seems quite limited and insecure for internal enterprise apps. @pgahq would be more likely to adopt this if you would enable the use of other Identity Providers (e.g., Okta) instead of Github authentication.

[agim]

What coding platforms does spark use to build apps?

I asked it to use rails and it decided to use react.

[dahatake]

I really want to attach functional requirement document at the first phase. Today we can attach only image to UX/Data prototype.
After initial app is up, I clone to repo to keep work colleague include GitHub Copilot.

[MohammedSardarSaajit4488]

Waiting ::))

[mhitza]

Needs a bit more time in the oven. Gave it initial prompt (asked react native, generated some kind of plain react application), didn't implement all the features requested. Asked it to fix a bug, then to implement a separate feature and I got hit with a too many requests error (429). These where 3 requests over the span of what I would call an hour (prompted and went afk for a while each time).

[himanshu2s]

🔧 Feedback on GitHub Copilot Spark Completion

📌 Subject:

Inaccurate and incomplete completions despite clear and structured prompts

---

🧠 Context:

The backend API is fully operational at [http://localhost:59756/swagger/index.html] and MongoDB is running at [mongodb://localhost:27017]. The frontend (React) was instructed to:

1. Remove all mock data and use the working backend endpoints.
2. Fix the subscription button visibility and interaction bugs.

However, Spark repeatedly responded with messages like:

"All tasks are complete", "Everything is resolved", or "System is working perfectly" — despite nothing being fixed or functional.

---

❌ Core Issues Observed:

• 🔁 False Completion Loops
  Spark asserts success even though the UI remains broken, mock data persists, and real API is not being called.

• 🧩 Lack of Real Integration
  It failed to replace mock logic or wire API calls into the existing UI components.

• 🧪 No Functional Validation
  There was no change in behavior — backend calls were not triggered, nor were frontend bugs fixed.

---

💡 Suggested Improvements:

1. Stateful Prompt Awareness
   Spark should evolve the code based on past prompts and the working system context (like functioning backend endpoints).

2. Backend–Frontend Integration Logic
   When both parts are present, Spark should intelligently connect them, using fetch/axios properly and handling responses.

3. Accurate Task Completion Messaging
   Avoid confirming completion unless issues are demonstrably fixed (e.g., API hits succeed, buttons render properly).

---

✅ Desired Outcome:

A more reliable Spark experience that:

• Replaces mock data with actual API usage,
• Fixes UI rendering bugs (like the disappearing button),
• Connects frontend components to backend endpoints seamlessly,
• Aligns environment configs correctly (CORS, ports, base URLs),
• And avoids misleading "success" messages.

---

Please let me know if code snippets, repo access, or error screenshots are needed for deeper review.