Let me be honest with you.
After I posted about IIT Madras wanting to talk to me on November 3rd, I was buzzing. I was ready to go. I had my asciinema demo queued up, my CTO Akhouri was locked in, and I was staring at my inbox every 5 minutes.
But Yamini from IIT Madras Incubation Cell hasn't replied yet.
So what did I do for 4 days? I hung out with my family in Tirupur, ate some amazing food, and then... I got bored.
Really bored.
And when a founder gets bored, they do what founders do best: they ship.
Over the last 48 hours, I went back into Termux on my $150 Android phone and turned HYNA-Stage from a "CLI tool that deploys sandboxes" into a complete, enterprise-grade Platform as a Service.
Welcome to HYNA-Stage v2.5.0.
π WHAT SHIPPED IN 48 HOURS
1. Native Push Notifications (VAPID + Service Worker)
No more third-party notification services. I built a native PWA implementation using @negrel/webpush with VAPID JWK keys. Now when you deploy a sandbox, you get a real, Vercel-style notification directly on your phone's lock screen.
2. The @hynaweb/sdk (Open-Source JavaScript SDK)
HYNA-Stage is no longer just a CLI. Developers can now programmatically deploy, upload files, and destroy environments using our open-source SDK. We are officially a platform.
3. Team Workspaces (Multi-Tenant)
You can now invite team members by username and share environment visibility. This required wrestling with Supabase RLS infinite recursion (solved with SECURITY DEFINER helper functions β ask me about this war story sometime π
).
4. Stateful Sandbox Snapshots
Using E2B's betaPause, your sandboxes can now be saved and resumed. Your state persists. Your work doesn't vanish when the session ends.
5. AI Template Generator
Integrated Groq's openai/gpt-oss-120b to generate valid template JSON from natural language prompts. Describe what you want, and the AI builds a validated, auto-tested template for the Marketplace.
6. The Dynamic Island UI
A draggable, mobile-first floating navigation bar with smart click/drag differentiation. Built specifically for small phone screens.
7. hyna-stage doctor + /dev/tcp Polling
Automatic CLI diagnostics, plus I replaced external Supabase polling with internal bash /dev/tcp polling. Deployments now complete in under 10 seconds.
βοΈ THE VERCEL COMPARISON (Yes, I Said It)
I know this sounds bold, but let's look at the feature parity:
| Feature | Vercel | HYNA-Stage v2.5.0 |
|---|---|---|
| CLI | β | β |
| SDK | β | β |
| Templates | β | β + AI-Generated |
| Team Workspaces | β | β |
| Notifications | Email/Slack | Native Web Push |
| Environments | Shared Serverless | Isolated Firecracker MicroVMs |
| State | Stateless | Stateful Snapshots |
| AI Support | Standard Chat | HITL with [HUMAN_REVIEW] |
Vercel is incredible. They are the gold standard. But HYNA-Stage is now playing in the same arena, with a few unique advantages (isolated MicroVMs, stateful snapshots, native push) that even they don't have.
And we did it all on a βΉ0 budget from a $150 Android phone.
π§ THE REAL LESSON
The last 4 days taught me something important.
Waiting is not wasted time.
While I was waiting for IIT Madras to reply, I could have just sat there refreshing my inbox. Instead, I shipped 7 major features. I solved RLS recursion. I built native push notifications. I wrote an SDK.
When that call finally happens, I'm not going to show them the same product I showed them on November 3rd. I'm going to show them a completely different platform.
That is the founder mindset. You don't wait for permission to level up. You level up while you wait.
π THANK YOU
To everyone who commented on the IIT Madras article, to Akhouri for being an incredible CTO, to Nicky for the elite CMO work, and to this community for believing in a 12-year-old with a $150 phone:
This v2.5.0 is for you.
The IIT Madras call is coming. And when it does, we are going to be ready.
Until then, keep building. Keep shipping. And if you ever get bored... build something.
Try HYNA-Stage:
npm install -g @hynaweb/stage
hyna-stage init node
π―βοΈ
Top comments (7)
That's the mentality! Most get stuck trying to make the perfect decision when most are reversable. It's better to fail fast than get stuck in indecission.
I want to point something out, to caution you against. Any sole AI connection added also adds a single point of failure. It would be good to add a way to have a "backup" integration and/or let the user change to something they'd prefer. I'm not going to pretend to know what you have going on under the hood - just a broad concern to keep in mind as you build. π€
Shipping seven major features in 48 hours from a phone is an impressive way to spend a waiting period β the stateful sandbox snapshots stood out most, since losing work when a session ends is exactly the kind of friction that makes a platform feel temporary. The Supabase RLS recursion fix sounds like the hardest part of the team workspaces work; was that the piece that took the longest, or did the native push notifications with VAPID take more debugging? Good luck with the IIT Madras call.
Great question, and thank you for noticing the snapshots! That was a massive quality-of-life upgrade for us.
To answer your question: the native VAPID push notifications absolutely took the most debugging. π
The Supabase RLS recursion was a fierce 2-hour battle, but once I wrapped my head around the
SECURITY DEFINERhelper pattern, it clicked. VAPID, on the other hand, was a 6-hour rabbit hole. Getting the JWK key formats to play nicely with Deno Edge Functions, ensuring the Service Worker registered correctly on mobile browsers, and handling the VAPID payload encryption was a massive puzzle.But seeing that first native lock-screen notification hit my phone made every minute of debugging worth it.
Thanks for the kind words and the good luck wishes for the IIT Madras call! π―βοΈ
Building a Vercel-level PaaS on a $150 phone via Termux is absolute madness! π₯ Shipping stateful snapshots and fixing RLS recursion in just 4 days is mind-blowing. IIT Madras will definitely be impressed. Keep shipping! π
Thank you so much, Badal! π
"Absolute madness" is exactly how it felt at 2:00 AM trying to get those VAPID keys to parse in Deno! π But you hit the nail on the head: fixing that RLS recursion and getting stateful snapshots working was the key to making this feel like a real, professional platform instead of just a temporary toy.
The goal has always been to prove that enterprise-grade infrastructure doesn't require a $3,000 MacBook. If it can be built on a $150 phone in Termux, it can be built by anyone with enough obsession and resourcefulness.
I really appreciate the encouragement for the IIT Madras call. Fingers crossed, and I'll be sure to share how it goes with the community! π―βοΈπ
HYNA-Stage?
Hey! π
HYNA-Stage is a platform that lets you spin up isolated, temporary cloud development environments (real MicroVMs, not shared "ghost accounts") in under 10 seconds.
You can do it directly from your terminal or our web dashboard. Itβs built to make cloud deployment accessible to everyone, even if you're just coding on a budget phone.
Check it out here: hyna-stagetemp.netlify.app/marketp...
Let me know if you have any questions or want to see how it works! π