About OnDevice Tools

Why this site exists

Some online utilities upload files for processing. Others work locally. It can be difficult to tell which you are using before choosing a file. This project makes its processing code public so you can inspect that choice and understand each tool's limits.

OnDevice Tools was built to remove that decision. Every tool here is written in JavaScript that runs on your own machine. Your file is read by the browser, processed by the browser, and handed back by the browser. There is no upload step to opt out of, in the tool-processing code. The public website still connects to its CDN, analytics and advertising providers; local processing is not a guarantee against every privacy risk.

How it actually works

The site is a static build. Each of the 64 tools is prerendered to plain HTML at build time and served as a file from a CDN — the same way an image is served. When you open a tool, the browser downloads a small JavaScript bundle for that one tool and runs it locally.

The heavy lifting uses capabilities browsers already ship: the Canvas API for image compression and resizing, the File API for reading files you choose, Web Crypto for hashing, pdf-lib for the PDF tools, and the built-in JSON and text parsers for the developer tools. Nothing about that pipeline involves a network request.

A service worker caches application resources for offline use. Open the tools you need and test them offline before relying on them without a connection. Downloads can fail, storage can be cleared, and a first visit does not guarantee a complete cache.

What this means in practice

• No processing upload. Speed depends on your device and input size.
• Use non-sensitive samples. Do not paste live credentials or regulated data into a public website solely because it processes inputs locally.
• Limits are tool-specific. Some tools cap input size or execution time to protect responsiveness; device memory also matters.
• It works offline once cached, including on a plane or a train.
• There is no account. Nothing to sign up for, nothing to cancel.

Who builds it

OnDevice Tools is built and maintained by Priyabrata Saha, an independent developer. It started in 2026 as a handful of tools that solved problems I kept hitting, and grew as I got tired of pasting things into sites I did not trust.

It is a one-person project, which is why tools ship steadily rather than all at once, and why the feedback loop is short: if something is wrong, mail support@ondevice-tools.org with a non-sensitive example and your browser version. Response and repair times are not guaranteed.

How it is paid for

Serving static files from a CDN costs very little, which is the main reason this can be free without an account or subscription. Donations support the project; advertising is configured but whether ads appear depends on provider approval and visitor settings. Our tool code processes inputs locally and does not send them to advertising integrations. Advertising scripts still make network requests; local processing does not isolate the page from third-party scripts.

Exactly what advertising does and does not have access to is set out on the privacy page, in plain language.

If you would rather support the site directly than see ads, you can buy me a coffee. It is genuinely optional: there is no membership, no ad-free tier and nothing withheld from anyone who does not.

You can read the code

The whole site is open source, on GitHub, under the AGPL-3.0 licence. That matters more here than it would on most projects: the central claim is about what this software does not do, and a claim like that is worth more when you can check it than when you are asked to take it on trust.

To inspect a particular workflow, use synthetic input and watch the Network tab in your browser's developer tools. Distinguish application downloads, analytics and ad requests from tool processing. Test that workflow offline too. These checks provide evidence about that session, not a security guarantee for future code or third-party scripts.

The licence is deliberate. AGPL-3.0 lets anyone run, study, modify and self-host this, commercially included — but if someone modifies it and offers it to others over a network, they have to publish their changes too. A permissive licence would let a modified, closed copy be run as a competing service with nothing given back.

What is not here

No account system, no cloud storage, no sync between devices, no team features and no paid tier. Those would all require a backend, and the absence of a backend is the entire point. If you need a tool that remembers your work across machines, this is the wrong site — and that is a deliberate trade.

Get in touch

Bug reports and tool requests go to support@ondevice-tools.org, anything else to contact@ondevice-tools.org — see the contact page for which is which. Requests for a tool that would need a server get an honest no rather than a maybe.