I Built a Zero-Login Postman Alternative in 5 Weeks. My Cofounder Is an AI and I Work Long Shifts.
Source: Dev.to
Why This Exists
One night I needed to test an API endpoint. I opened Postman. It asked me to:
- Create an account.
- Download a desktop app.
- Set up a workspace.
- Pick a team plan.
I just wanted to send one POST request.
I thought: somebody should build the version of this that just works when you open it—no download, no login, no workspace wizard. Open a browser tab, describe what you need in plain English, and get a production‑ready API call in seconds. Hit Send. That’s it.
I built it. You can try it right now with zero login at .
I didn’t know how to build it. That turned out to matter less than I expected.
Five Weekends
Tools: Claude running inside Cursor, plus whatever hours I could carve out between Friday and Sunday.
Workflow:
- Weekdays: I work my regular job.
- Weekends: I build. The AI writes the implementation; I write detailed specifications, review every line, test the output, and own what ships.
Stack: Next.js 16, React 19, TypeScript in strict mode, Tailwind CSS v4. I chose these because Claude recommended them and the reasoning was sound. I didn’t fully understand TypeScript strict mode at the start, but the compiler rejected nearly everything I shipped in the first week, forcing me to learn.
Week 1 – Core Request Builder
- URL input
- HTTP‑method selector
- Headers & body fields
- Send button
- Formatted JSON response
The first time I sent a real GET request to a test API and saw structured data return in something I built, the project became real. Everything after that was momentum.
Week 2 – Save & Collections
- Persisted to
localStorage(no database, no accounts, no server‑side storage). - Your data lives in your browser and never touches my servers.
I made this decision because I didn’t know how to build a database layer. It turned out to be the best architectural choice in the entire project: privacy by default, not by policy.
Week 3 – Environments & Variable Substitution
- Define
{{API_KEY}}or{{BASE_URL}}once. - Switch between Dev and Prod instantly; every URL, header, and body updates automatically.
The regex was simple, but the state‑management to sync multiple environments with an editor panel while persisting everything to localStorage was not. The AI generated something that looked correct but had a subtle bug—closing and reopening the panel was the only way to see updated variables after switching environments. I spent an entire Saturday finding it; the fix was three lines.
That same week, I asked the AI to update a documentation file and it silently modified a completely different component at the same time. I only caught it because something looked different in the browser. Lesson: AI follows your instructions precisely and sometimes does exactly what you didn’t ask for at the same time.
Week 4 – Stripe Payments, JWT Auth, & Free Tier
Pro tier ($12 / month): straightforward.
Free tier: wanted to give 3 AI generations per month without any login.
Solution: use Upstash Redis to track a SHA‑256 hash of a composite fingerprint—IP address + User‑Agent + Accept‑Language headers. No PII, no cookies, no accounts. Just a hash that tells the server this browser has one generation remaining this month.
Pro users: magic‑link authentication (single‑use tokens, 15‑minute expiry stored in Redis), Stripe subscription verification on every link click, rate limiting to prevent abuse. No passwords, no traditional accounts—just an email and a secure link.
The AI generation itself knows 50+ real APIs (GitHub, Stripe, OpenAI, Twilio, Slack, Discord, Notion, etc.). Example:
“Create a POST request to Stripe to create a customer”
The AI builds the full request with the real endpoint, correct headers, proper auth pattern, and a {{STRIPE_SECRET_KEY}} placeholder ready for your environment variables.
You can try it at —three free generations, no signup.
Week 5 – Deployment, Custom Domain, Landing Page
- Live at apibuilderhq.com.
- Monthly infrastructure cost: ≈ $23.
What Actually Went Wrong
The sections above make this sound smooth. It was not.
Silent Vercel Deploy Failures
I’d make changes, run npm run build, get a clean pass, push to GitHub, and nothing would happen. No error, no deployment—just silence. After a long, embarrassing investigation I discovered Vercel was silently blocking every push because the Git commit identity didn’t match what it expected. The build passed, the push succeeded, but the deployment never triggered.
Fix: create a deploy hook—a curl command I now run manually after every single push. I still run it today, every time, without exception.
Environment‑Variable Typo
A typo in an environment variable—apibuildhq.com instead of apibuilderhq.com—was silently redirecting paying customers to a 404 page after checkout. Two missing letters, hours of debugging.
Takeaways
- AI can write code that compiles, but you still need to understand, test, and own the result.
- Privacy‑by‑default can be achieved with clever client‑side storage; you don’t need a full backend to start.
- Deploy pipelines can fail silently; always verify that your CI/CD actually triggers.
- Small mistakes (a missing “r”) can have outsized impact; double‑check everything that goes to production.
If you’re skeptical about building something real on weekends with only an AI assistant, give it a try. The journey is messy, but the learning—and the product—are worth it.
# Type‑checks, and passes every lint rule
It cannot tell you that your deployment pipeline is silently broken or that you misspelled your own domain name. The gap between code that works and a product that ships is entirely human.
---
## What I Learned About Building With AI
People want to know if *vibe coding* is real. Here's what I can say after five weeks.
- The AI is excellent at boilerplate, type definitions, consistent patterns, and generating validation logic that would take me hours to write by hand.
- It catches things I don’t know enough to think about — accessibility attributes, error‑handling patterns, TypeScript strict‑mode requirements, edge cases in data parsing.
**What it isn’t:** a replacement for understanding what you’re building. Every week there was a moment when the AI produced something that looked right but wasn’t. Every fix required me to read the code, trace the logic, and figure out why it was wrong. The AI made me faster, but it didn’t make me less responsible.
> *If you’re thinking about building something and you’re not sure you’re qualified — I wasn’t either. The qualification is willingness to sit with confusion until it breaks. It always breaks eventually.*
---
## Where It Stands
- **The product is live.** Revenue: **$0**. Users: **early**.
- **The builder is free.** Send requests, save collections, manage environments with variable substitution — no limits, no login, no catch.
- **Pro tier:** unlimited AI generation for **$12/month**. If the free version is enough, use it forever. That’s by design.
I’m not asking you to support a founder. I’m asking you to try a tool and tell me if it’s useful enough to keep open in a tab. The answer might be **no** — that’s useful too.
---
## One Last Thing
I want to test something.
**What’s the most undocumented, frustrating API you’ve had to integrate lately?** Drop the name below — I want to see if **APIBuilderHQ** can write the request in one shot.
---
*Built with Next.js 16, React 19, TypeScript strict mode, Tailwind CSS v4, Upstash Redis, Stripe, and Claude in Cursor.*
*[@APIBuilderHQ on X](https://x.com/APIBuilderHQ) · [apibuilderhq.com](https://apibuilderhq.com/)*