SETOOLS
Your private, on-device toolkit — nothing leaves your device. Included.
Read-aloud
Paste text (or open a .txt) and have it read to you — entirely on your device. No audio or text leaves your browser.
Voices come from your device/browser and vary by OS — pick one, adjust speed & pitch (your choice is remembered on this device). The word being read is highlighted and scrolls into view. “Read from image” uses the on-device OCR to pull text from a picture — nothing is uploaded. Dyslexia view adds spacing, a calm background and a dyslexia-friendly font.
Translate
🔒 On-device machine translation — nothing leaves your device. Your text is translated entirely in your browser and is never uploaded. The first translation for a language pair downloads its model to your browser once (~20 MB). A smaller language set than cloud tools — that's the privacy trade. Verify important text.
This release translates between English and Spanish, French, or German (English on one side). More language pairs are added later. Translation works on your own text — SETOOLS does not generate documents.
Voice — speech to text
🔒 On-device transcription — nothing leaves your device. Your audio — a file you pick or your microphone — is decoded and transcribed entirely in your browser by a self-hosted Whisper model, and is never uploaded; neither is the resulting text. The first time you use a model it downloads to your browser once (about 80 MB, plus a one-time ~22 MB engine). Accuracy is good but not perfect, especially with accents, noise or crosstalk — check important transcripts. This turns your own speech into text; it does not translate and does not generate.
Microphone recording asks your browser for mic permission and stays on your device. Live “Dictation (beta)” re-transcribes the last few seconds as you speak — best-effort, and it may lag; the dependable path is record-a-clip → Stop & transcribe. You can also download the transcript as a timestamped .srt. Whisper’s translate-to-English mode is intentionally not offered here — use the Translate tool for translation.
Background removal
🔒 On-device — nothing leaves your device. Your image is segmented entirely in your browser by a self-hosted model and is never uploaded. The first use downloads the engine + model to your browser once (about 31 MB). With rails: this only removes the background or replaces it with a solid colour, a blur of your original, or your own image — it never invents or generates pixels. Works best on a clear subject; check the edges.
Segmentation only — SETOOLS edits your own image and never generates backgrounds (no AI / generative fill). Model: U²-Net (Apache-2.0), run on-device via onnxruntime-web (MIT). Large images are scaled to a reasonable size for output.
Quick Sign
🔒 On-device fill & sign — nothing leaves your device. Add your own signature to a PDF or image in your browser and download a flattened copy. Your document is never uploaded.
Free fill & sign — it places a signature image and flattens the file. It is not a verified or legal-grade signature: no identity check, no audit trail, no tamper-evidence. Drag the signature on the page to move it; use the sliders to size and rotate. Reuses SETOOLS' on-device PDF tools (pdf-lib / pdf.js); nothing is uploaded.
PDF tools
Merge, split, rotate, reorder/delete pages, watermark, compress, and convert between PDF and images — all in your browser.
PDF wave 1b — all on your device. (Protect / Unlock above are now live and use the password box.)
🔒 True redaction: a page you black out is flattened to an image with the selected areas burned in — the underlying text and objects are removed and cannot be recovered or extracted. Pages you don't redact keep their text. Irreversible — keep your original.
Office↔PDF conversion is deferred — a faithful converter would need a LibreOffice-class engine (hundreds of MB), and SETOOLS will never upload your file to a server to convert it.
OCR — extract text from a scan or photo
Pull the text out of an image of your own document, on your device. The first run downloads a small language model to your browser (one time); the image itself is never uploaded.
Document scan
Capture or pick a photo of a document, enhance it for legibility, optionally crop, and save it as a PDF — all on your device.
Manual crop & rotate ship in wave 1b. Automatic edge-detection / dewarp is still deferred — no clean, small on-device approach yet (it would need a heavy CV/wasm dependency).
QR code generator
Turn a link, text, Wi-Fi login or contact card into a QR code — generated on your device and saved as a PNG. Nothing is uploaded.
Image compress / resize
Shrink or resize your own photo in the browser — pick a maximum size and quality, then download the result. The image never leaves your device.
Metadata / EXIF stripper
Remove embedded metadata (camera info, GPS location, timestamps) from your own image by re-saving it on your device. Download a clean copy — nothing is uploaded. Note: re-saving re-encodes the image — JPEGs are slightly recompressed; PNGs stay lossless.
Word & readability count
Paste text or open a .txt file to get word, character and sentence counts plus a readability score — calculated entirely on your device.