Developer tools that work offline, and how to check they really do
· 8 min read
Most of the small utilities developers reach for every day — format this JSON, decode this token, preview this README, generate a password — are a search away. Most of them also send whatever you paste to a server you know nothing about. That is a strange default for tools that routinely handle API responses with customer data, production config files and authentication tokens, especially when every one of those jobs is simple enough for the browser to do on its own.
"Works offline" has become a selling point for exactly this reason. It is also a claim anyone can make, so this guide covers both halves: which tools can genuinely run with the network off, and how to check that one really does.
What "works offline" can mean
The phrase covers three quite different things:
- Processes locally, but needs the network. The work happens in your browser, but the page breaks the moment you disconnect. Better than an upload, but you are trusting the claim.
- Keeps working after it has loaded. Close the connection with the page open and it still works, until you reload.
- Opens with no connection at all. The site caches itself (usually with a service worker), so after one visit it opens and runs on a plane, in a tunnel, or on a locked-down corporate network.
Only the third is offline in the sense most people mean. It is also the strongest evidence that nothing is being uploaded: a tool that works with the network physically unavailable cannot be sending your data anywhere.
A two-minute test for any tool
You don't have to take anyone's word for it, including ours.
1. Watch the Network panel
Open the browser's developer tools (F12, or ⌥⌘I on a Mac), switch to Network, and clear the list. Now use the tool: paste your JSON, drop your file, press the button. Every request the page makes appears in that list. A tool that processes locally adds nothing when you use it. If a request appears at the moment you hit "Format", click it and look at its payload — that is your data leaving.
Filter by Fetch/XHR and also look at WS (WebSockets), because a socket opened at page load can carry data later without showing up as new requests.
2. Pull the plug
In the Network panel, change the throttling dropdown from No throttling to Offline — or simply turn on airplane mode. Then use the tool again. If it still works, the work is happening on your machine. If you want to test level 3, reload the page while offline and see whether it opens at all.
3. Read the security policy
In the Network panel, click the page itself and look at its response headers for Content-Security-Policy. A connect-src 'self' directive means the browser itself refuses any script on that page trying to contact another domain. That is a promise enforced by the browser rather than by the site's good intentions. It is how this site is built.
Being straightforward about what you will see if you run this test on KeepItLocally: one request to /_vercel/insights on our own domain when a page loads, which is an anonymous page-view count. It contains the page address, never anything you paste or drop into a tool, and the privacy policy lists exactly what it records.
Tools that run fully offline
Everything below runs in your browser and, after you have opened its page once while online, opens and works with no connection at all — the site's service worker keeps a copy of each page you visit.
Data and code
- JSON formatter & validator — pretty-print, minify, sort keys, and for most syntax errors the exact line and column. API responses are the classic thing not to paste into a stranger's website.
- JSON ↔ YAML ↔ TOML converter — for config files, which tend to contain hostnames, connection strings and the occasional secret.
- Markdown preview — live GitHub-style rendering with tables and task lists, for READMEs and docs you are not ready to publish.
- Diff checker, regex tester, SQL formatter and code formatter — the everyday ones.
Tokens, hashes and encodings
- JWT decoder & verifier — a live token is a credential. Decoding one on a website hands that credential over; see what a JWT is for why.
- Hash generator — MD5 and the SHA family for text or whole files, with no upload of multi-gigabyte downloads just to checksum them.
- Base64, URL encoder and HTML entities.
- Unix timestamp converter and cron parser.
Generators
- Password generator — the one tool where "offline" matters most. A password generated by a server existed somewhere other than your screen before you ever saw it. Generated locally with the browser's cryptographic random source, it never did.
- UUID, ULID and Nano ID generator.
- QR code generator — free, with SVG export for print and no tracking redirect inside the code. More on that in QR codes that never expire.
Everyday
- Stopwatch, timer and Pomodoro — an offline stopwatch sounds like the least sensitive tool imaginable, and it is; it is just useful to have one that works on a train.
- PDF tools — merge, split, compress, password-protect and more — which is where avoiding uploads matters most of all.
Two caveats worth knowing
- Open each page once while online. A service worker can only serve what it has already stored. Visiting the tools you rely on once, on a good connection, is what makes them available later on a bad one.
- Some tools fetch an engine on first use. OCR, for example, downloads its language model the first time it runs and caches it afterwards. The download is a one-way fetch of a model file; your document is still never sent.
Why this is worth caring about
Pasting data into a web tool feels like using a calculator. Technically it is often closer to emailing that data to a stranger who promises to delete it. For most snippets that doesn't matter; for tokens, credentials, customer records and unreleased work it does, and it is exactly the kind of leak that never shows up until it is a problem. Tools that work with the network off take the question off the table — and the two-minute test above lets you confirm it instead of trusting a badge on a website.