Nanofile

Is the client open source?

Doing it locally is the part I like. What's the tradeoff though, is it slower on big files, or is there a size where the browser just gives up? Trying to work out whether "never uploads" costs anything in practice.

Nanofile Your product has strong potential, but I found a few key improvements that could make it even better. I'd love to share my feedback and suggestions—please contact me at [email protected].

Sowiks does automatic compression too, but server-side after upload, so "never leaves the device" is the part I'm most curious about here — in-browser codecs can get memory-hungry fast once you're batching a whole folder. Have you hit real ceilings on mobile Safari specifically, or has that mostly been a non-issue outside desktop Chrome? Detecting format by magic bytes instead of trusting the extension is a nice touch too — that catches the mislabeled-file case that trips up a lot of similar tools.

The "without uploading them anywhere" part is the real differentiator for me — most compressors quietly round-trip your files through a server. Curious what you're using under the hood for the ~68% ratio: is it format-aware (re-encoding images/PDFs differently) or one general-purpose algorithm? And is there a file-size ceiling where the browser tab starts to struggle?

Magic-byte detection instead of trusting the extension is the detail that stands out to me — that's exactly the kind of edge case that's easy to skip when you're moving fast. I do something similar with format handling on calculate.at's converters, and it's always the mislabeled or malformed inputs that eat the most debugging time, not the clean ones. With 37 formats in one drop zone, was there one pair that turned out way harder to disambiguate than expected?

The "no upload endpoint, nothing to intercept" pitch is exactly why we keep TidyCalcs' calculators client-side too — it's a real trust advantage once people notice it. With 37 formats routed through magic-byte detection and batch processing across a whole folder, how are you keeping the drop zone fast — single worker thread for the whole pipeline, or parallelizing across files?

The thing nobody warns you about with client side only is support. I ship a local-first desktop app and the flip side of never seeing user data is that when someone tells you it produced a broken file, you have nothing. No logs, no repro, just a stranger's word and a file you cannot look at. With 37 formats in play the failure modes must be wildly different between STL, HEIC and DOCX. Do people end up mailing you the input file, or do you mostly only fix the ones you can reproduce yourself?