51 downloads in 2 days (some maybe bots). Built a waitlist page anyway.
Two days ago I shipped @keelstack/guard – a tiny TypeScript package that prevents AI agents from calling the same tool twice.
No marketing. No launch. Just npm publish and a README.
The numbers so far:
- 51 downloads on npm
- 1 stars (mine) on GitHub
- 0 issues opened
- 0 DMs saying "thank you" or "this sucks"
GitHub shows 108 unique cloners and 293 clones. Some of those are definitely bots (npm mirrors, CI systems, scrapers). But some of them might be real developers who hit the same duplicate tool call problem I did.
That's enough signal for me to keep going.
The problem I'm solving:
AI agents retry tool calls when things timeout or fail. Your email service gets called twice. Stripe gets charged twice. Your database gets duplicate writes. No agent framework solves this natively.
'@keelstack/guard' wraps any async action with idempotency, budget limits, and risk gates. MIT. Zero deps. 37 tests, 98% coverage.
What I built this week:
A waitlist page for the upcoming dashboard + Redis adapter + OpenTelemetry spans.
No fake numbers. I literally put "0/100 founding members" on the page.
What's next (planning):
A Chrome extension that shows real-time costs, API calls, and duplicate call detection – right in your browser while you're building. Think of it as a live debugger for your AI agent's tool calls. Still early, but I think developers need visibility into what's actually happening when their agent runs.
The offer for early people:
First 100 founding members lock in their base plan price forever. When I raise prices later (and I will), your price stays the same.
No credit card. Just email.
Why I'm posting this:
I'm 17. I've been building in silence for weeks (maybe months). The backend engine I originally built (auth, billing, queues) got zero paying customers. This pivot feels different – the problem is sharper, the package is smaller, and 51 downloads in two weeks (even with bots) is more traction than the engine ever got.
I need feedback. I need people to break this thing. I need to know if I'm solving a real problem or just my own.
The page: https://guard.keelstack.me
The GitHub: https://github.com/KeelStack-me/guard
The npm: npm i @keelstack/guard
If you've ever had an AI agent send the same email twice or charge a customer twice, I'd love to hear your story. If you think this is useless, tell me that too.
Building in public means showing the good and the ugly. 51 downloads is not a win yet. But it's not zero.
The interesting part is not the raw download count, it is that the package is sharp enough to earn curiosity without any launch motion. If that keeps happening, the bigger opportunity might be using the open-source package as the trust layer and the dashboard/visibility tooling as the paid layer.
You're right, the downloads are noise, but the fact that people get the problem without me explaining it is the real signal. I'll keep the core free MIT and focus paid on the visibility layer (dashboard, Redis, OTel). Thank you, this is the clarity I needed.
I know a few developers working with AI agents who have run into this exact problem. I'd be happy to ask if they'd answer some questions for you if you'd like to put some together.
Preventing AI agents from double-charging Stripe or spamming emails is a mission-critical problem, Siddhant. Moving from a bloated "backend engine" to a hyper-focused, zero-dependency TypeScript package is exactly how a 17-year-old developer out-builds enterprise teams—by solving a sharp, expensive pain point.
I’m currently running a project in Tokyo (Tokyo Lore) that highlights high-utility logic and young, resilient builders like you. Since you've pivoted from "building in silence" to shipping a live idempotency guard for AI agents, entering your project could be the perfect way to lock in your "founding member" status with a global network while your odds are at their absolute peak.