The tools I decided not to build

Every month I get email from Toolium users asking why the site does not have a particular tool. The requests vary, but the most common are OCR for scanned documents, AI image upscaling, currency conversion, video format conversion, audio editing, and custom font generation.

I have thought about each one. None of them are on the roadmap. This post is why, because feature restraint is probably the most contested design decision I have made.

The principle

Toolium's design constraint is straightforward. If a tool cannot be done well in a browser tab, it does not ship. "Well" means the result is as good as a competent desktop equivalent, and the user experience does not require significant patience.

Most of the "missing" tools fail one or both of those tests. They are technically buildable in a browser. The implementations would be visibly worse than what users could get from a desktop app or a paid service. Shipping a worse version of a thing under a brand that is built on quality undermines everything else the site is doing.

OCR

Tesseract.js exists. It works. The English-language model is 10 megabytes. Recognition takes 30 to 60 seconds per page on a typical laptop. Accuracy on real-world scans is around 90% for clean documents, lower for poor scans.

Ninety percent accuracy means a thousand-word document has a hundred wrong words. For a contract or invoice, that is worse than useless - it gives the appearance of digitized text while corrupting most of the information. Compare to ABBYY FineReader at 99% on the same inputs. Or to Tesseract on a desktop with the larger language models loaded, also significantly higher than the browser version.

If Toolium shipped OCR at 90% accuracy, the support load from "the output is wrong" would be substantial. If I waited until I could match FineReader in a browser tab, I would be waiting forever.

Decision: no OCR until the open-source models close the accuracy gap and the inference time drops. Both are plausible within a few years. Neither is here yet.

AI image upscaling

The model behind tools like Topaz Gigapixel is Real-ESRGAN or a close cousin. The open-source version is around 20MB to download. Running it on a CPU (no GPU acceleration) takes 30 to 60 seconds per image. The output quality is excellent. The UX of waiting through 30 seconds of GPU-less inference in a browser tab is bad enough that Toolium would feel broken.

WebGPU could help. The browser-native GPU API is now stable in Chrome and Edge. Safari progress in 2025 was encouraging. Once WebGPU is universal, AI upscaling becomes plausible at reasonable speed.

Decision: revisit when WebGPU works on every major browser. Until then, the desktop tools and paid services have a clear UX advantage.

Currency conversion

The data is fundamentally external. Exchange rates change every minute. Toolium has no backend; the rates would have to come from a third-party API on every page load, which would add a network request, latency, and a reliability dependency for what is essentially a one-line lookup.

The honest path is to use xe.com or Google. They have the rates. They update them. Toolium would just be a thinner wrapper over the same underlying data.

Decision: no currency conversion. Physical units (length, weight, volume, temperature, data size) are the tractable case, because the conversion ratios are mathematical constants and never change.

Video format conversion

ffmpeg.wasm exists and works. It is 30 megabytes to load. Encoding a 10-minute video takes 5 to 10 minutes of CPU time in a tab. The use case for "I just want to convert this MP4 to MOV right now" has a much better answer than "wait ten minutes in a browser tab." That answer is HandBrake on the desktop. It is free, open-source, runs in seconds with hardware acceleration, and handles every format anyone needs.

Decision: no video tools until there is a hardware-accelerated path in browsers (WebCodecs is partially there but the encoder side is limited) or the file sizes drop dramatically. Neither is imminent.

Audio editing

The Web Audio API is powerful. AudioWorklets can do real-time DSP. The problem is that "audio editing" in the user-facing sense usually means waveform visualization, snipping, fading, multi-track arrangement - and the ergonomics of doing that in a browser are not yet competitive with Audacity (free, runs on every desktop) or Reaper (paid, professional-grade).

The right tool already exists, runs on every desktop, and is not a browser tab.

Decision: no audio tools. The category is well-served outside the browser.

Custom font generation

The tools that exist for this (FontForge, Glyphs, Birdfont) are mature, complex desktop apps designed for typography professionals. Building a worse version in a browser would be a category error - the workflow is too involved and the audience too narrow to fit Toolium's "everyday utility" thesis.

Decision: not a Toolium tool, ever, regardless of how the browser ecosystem evolves.

The meta-point

Each of these has a "but you could ship a version that works for 60% of cases" argument. I have heard them. The reason I do not is that 60%-of-cases tools have a specific failure mode: they look like they work, then they do not, and the user has wasted time and may not realize the output is bad. For a free tool with no support, "looks fine but is wrong" is the worst possible failure.

The tools Toolium does ship pass a higher bar. They either work correctly, or they refuse the input and explain why. The output of a Toolium PDF merge is bit-identical to what Acrobat would produce. The output of the Image Compressor is honest about the trade-off it made. The output of the Regex Tester is what the regex engine actually returned, not a best guess.

That bar is what keeping tools off the roadmap maintains.

What is on the roadmap

The next tools I am considering are smaller and more specific than the ones I declined.

An EXIF metadata stripper for photos. Useful for sharing photos without revealing location, camera serial number, or capture time. Built around the existing image pipeline; small library, no UX complications.

A markdown table generator. The kind of small utility that turns five minutes of fiddling into thirty seconds. Pure UI work, no heavy compute.

A more specialized JSON tool for API debugging. Show diffs between two responses, schema validation, path queries. The browser is the right place for it; the use case is well-defined.

If you have a request that hits a real pain point, the email is dev@devtoolium.com. The most useful requests come with a specific workflow. "I wish there was a tool for X" gets considered. "I tried Y to do Z and it failed because of W" gets prioritized.