I kept doing the same thing every few days: googling a Windows error code, pasting a JWT into some random site to decode it, or working out a CIDR range by hand because I couldn't remember the shortcut. Dozens of tabs, dozens of sites, none of them quite trustworthy with what I was pasting in. So I built one place for the tools I actually reach for: errno.help - CIDR/IPv4 calculator, JWT decoder, hash generator, UUID/password generator, YAML⇄JSON, cron parser, DNS/WHOIS/ASN/GeoIP lookups, a Windows error/event ID reference, and a few command builders (Robocopy, chmod, kubectl).
It's open source (MIT): github.com/ssakhavat/errno-help
A few decisions and mistakes along the way felt worth writing down.
Client-side first, server only when it can't be avoided
Most of these tools don't need a server at all. CIDR math, JWT decoding, hashing, UUID generation - all of it runs in the browser with native APIs (crypto.subtle.digest, crypto.getRandomValues, crypto.randomUUID). Nothing you paste into the JWT decoder ever leaves your machine, which matters more than it sounds like it should when the tool's whole job is decoding tokens.
A handful of tools genuinely need a server - DNS, WHOIS, ASN, GeoIP, and a port checker. Those go through Next.js API routes that talk to an upstream service, so no API key or raw socket access ever touches the client.
The GeoIP provider swap
I originally wired GeoIP lookups to MaxMind's GeoLite2 web service - free, well-documented, industry-standard data. I set up Basic Auth with an account ID and license key, deployed it, and got: "The GeoIP service rejected our credentials."
Locally, the exact same credentials worked fine, curl included. On Vercel, rejected, consistently, across multiple redeploys and freshly re-pasted environment variables. I never fully root-caused it - possibly an account-level web-service permission that hadn't propagated, possibly something in how the credentials moved through Vercel's env var UI. Rather than keep debugging a black box, I switched to ipinfo.io, which uses a single URL-embedded token instead of Basic Auth. Same result, less to get wrong, and - as a bonus - its free tier's terms are commercial-use friendly, whereas the always-free ip-api.com option I'd also considered is explicitly non-commercial only. Worth reading the fine print before you build on top of a free tier.
The rate limiter I thought I'd already built
Every network-facing route had rate limiting from early on - capped per IP, tighter on the port checker since it's the closest thing to a port scanner I'm willing to expose. Then I asked for a general security pass, and it turned up something I'd missed entirely: rate limiting keyed on X-Forwarded-For can be trivially bypassed by just sending a different fake value on every request. Twelve spoofed headers, twelve requests that should've been blocked at ten.
The fix ended up being environment-aware: on Vercel, the platform's own edge network sets x-real-ip in a way the client can't override, so that's authoritative there. Everywhere else (local dev, unknown environments), each client gets a random ID in an httpOnly cookie and its own isolated bucket, so spoofing gains nothing and legitimate users don't share a quota with strangers. It's the kind of bug that's invisible until someone actually tries to break it - and a good reminder to occasionally treat your own tool as adversarially as a stranger would.
One screen, no scroll, no logo
The design brief I gave myself was strict: single viewport, no logo, plain typography, nothing that looks templated. That constraint turned out to be harder to satisfy structurally than visually. A shared overflow: hidden; height: 100vh on <body> - meant only for the landing page - quietly clipped every other page in the app, including tool results that overflowed a single screen. A pre-existing media query meant to relax that on short viewports never fired, because a bare-element CSS selector can't out-specify a Tailwind utility class. The fix was moving the viewport lock off and onto the landing page's own root element, so it's opt-in per page instead of a global default nobody remembers is there.
What's next
The tool list will probably keep growing - Base64, regex testing, a few more command builders are on the list. If you build or maintain IT/dev tools yourself, I'd be curious what you reach for that isn't here yet.
