The short version
A modern browser can read a file from your disk, decode it, transform it and hand it back, without any of it leaving your machine. Tools built that way cannot leak your file, because there is no transfer to leak. You can verify the claim yourself in under a minute: open the Network tab and convert something, or open the page once and then disconnect from the internet and use it again.
Two different machines do the work
| Upload-based tool | Browser-based tool | |
|---|---|---|
| Where the file goes | Across the internet to a server | Nowhere. It stays on your disk and in your tab |
| Who can read it | The operator, anyone with server access, anyone who breaches them | You |
| What protects you | A promise in a policy | The absence of a transfer |
| Works offline | No | Yes, once you have opened that tool |
| File size limits | Set by the operator, for cost and safety | Set by your own memory |
| Queues and waiting | Yes, at busy times | No. Nobody else is in the queue |
| Can be verified | Not really | Yes, in under a minute |
None of this makes upload-based tools evil. Some jobs genuinely need a server. It makes them a choice you should be making deliberately rather than by default.
The last row is the one people meet first. A server's upload cap is set by the operator, for cost and safety, not by what your file needs, which is why so many people arrive here having just been told their file is too large to upload or that its type is not supported. Both are fixable on your own machine, before anything is sent.
Two ways to test any site, including this one
Do not take a site's word for it, including ours. Both of these work on any converter on the internet.
- Watch the network. Press F12 to open developer tools, choose the Network tab, clear it, then convert a file. Every request the page makes is listed. If your file is being uploaded you will see a request the size of your file, going out. On a browser-based tool you will see the page's own assets, and a small page-view count — nothing carrying your file.
- Work offline. Open the tool once so the browser has its files, then turn off your Wi-Fi or enable aeroplane mode and convert. A tool that needs a server fails at that point. One that keeps working is doing the work on your machine. It has to be a tool you have already opened: nothing can run from a page the browser has never downloaded.
The second test shows the work does not need a server. On its own it cannot prove nothing is sent later, because a page can queue a request for when the connection returns, so run the Network test as well.
What you will see going out, and why. This site counts page views, through Cloudflare's own analytics. The counter sets no cookies, stores nothing on your device, and does not follow you between visits or between sites — which is why there is no cookie banner to dismiss. It records that a page was opened, how quickly it loaded, and broad details such as country, browser, operating system, device type and the referring site. It does not collect the file you are working on: Cloudflare describes it as collecting page-load timing, and the file is never handed to the page's network code. If you are checking the Network tab, those are the requests you will find (the counter's script from static.cloudflareinsights.com and a small report to /cdn-cgi/rum); a request carrying your file would be roughly the size of your file, and there is not one.
The files worth being careful with
- Anything with a signature on it. Contracts, agreements, letters of offer.
- Identity documents. Passport scans, driving licences, visa paperwork. A copy in the wrong hands can be used to impersonate you.
- Money. Payslips, bank statements, tax returns, invoices with account details on them.
- Health. Test results, letters, insurance claims.
- Other people's data. A customer export, a spreadsheet of email addresses, a database dump. Uploading that is not just your risk to take.
- Credentials. A live token pasted into a formatter is a security incident, not a convenience.
- Photographs of your home. Which, as it happens, usually carry the coordinates of your home inside them.
What a browser still cannot do well
A page claiming a browser can do everything would be making the same kind of unchecked promise this one is arguing against. It cannot.
- Heavy video. Re-encoding a long video in a tab is possible in principle and painful in practice.
- Reading text from images. Optical character recognition needs a language model of several megabytes or more before it does anything at all.
- Removing backgrounds. Same problem: it is a machine learning model, and downloading it costs more than most people expect.
- Deep PDF surgery. Re-compressing every embedded image inside a PDF is far harder than re-saving the file, and a tool should say which of the two it does.
- Licensed codecs. Some formats are patent-encumbered and not built into every browser, so a page has to ship its own decoder where the licence allows, or cannot handle them at all.
Where a tool on this site cannot do something, the page says so.
117 tools, none of which upload anything
Images
HEIC, WebP, JPG and PNG in any direction. Compress, resize and strip EXIF.
Documents
Shrink, merge and split PDFs without uploading a contract.
Convert
Move data between JSON, CSV, YAML, XML, Markdown and more.
Developer
Format, decode, hash and test without pasting into a stranger.
Text
Count, clean, compare and reshape text at speed.
Calculators
Money, dates, health and grades, worked out in the tab.
Generators
Passwords, QR codes, UUIDs and test data, made locally.
Colour and design
Palettes, gradients, shadows and contrast checks.
Time
Timers, countdowns and working-day arithmetic.
Start with compressing a PDF, converting an iPhone photo, or the full list. Curious how the formats actually compare? The format benchmark measures your own files rather than quoting an average. If something has just refused a file of yours and you are not sure why, why was my file rejected works through every common reason, including attachment limits in email.
Frequently asked questions
Last reviewed 2 October 2026