SubtitleMatrix - 243 subtitle tools that never upload your files.

Hi Product Hunt — I built SubtitleMatrix because every subtitle tool I tried asked me to upload the file first. That requirement never made sense to me. A subtitle file is text. Most subtitle work — converting SRT to VTT, shifting timings, cleaning tags, checking reading speed — is a text transformation. It does not need a server, a queue or an account. Yet the category is full of upload boxes, because the code lives on someone else's machine. So I inverted the architecture: there is no backend. Not disabled, not opt-in — there is simply no endpoint to send a file to. Parsing, transformation and download all happen in the tab. That matters if you handle unreleased footage, legal or medical material, or anything under NDA. Open DevTools while a conversion runs and you will not see a single outbound byte carrying your file. What is in it: - 243 tools across 8 categories — format conversion, timing, text and styling, quality checks, accessibility compliance, timecode, export and media input. - Loss reports before you run anything. ASS → SRT drops styling, positioning and karaoke timing; the page tells you that up front instead of letting you discover it in the output. - Batch mode — drop a whole season, process it serially, download one zip. - Compliance checks against EN 301 549, WCAG 1.2.2, FCC, Ofcom, Netflix and BBC thresholds, with a PDF and CSV report that cites the clauses it applied. The part I am most proud of is the media category, because it sounds impossible without a server: extract subtitle tracks from video with ffmpeg.wasm, OCR PGS and VobSub bitmap subtitles, transcribe speech with Whisper, generate narration with Kokoro, and resync subtitles against an audio track. The engines are self-hosted files downloaded into the tab — the media never leaves your machine. First use downloads roughly 30–100 MB depending on the tool, and OCR or speech results should be proofread: they are best effort, not a guarantee. One thing I deliberately did not build: translation. Every machine-translation path requires sending the cue text somewhere. That would break the one promise the site is built on, so I left it out. It is free, with no account and no upload. Two things I would love feedback on: which format or operation is missing from your workflow, and whether the loss reports are specific enough to actually be useful. I will be around all day.