A command-line tool for macOS

Your Mac can see and hear.
Now your terminal can too.

OCR, detection, segmentation, transcription and speech — powered by Apple frameworks, entirely on device. No network. No cloud. Ever.

$ brew install wkoszek/tap/sense

Signed, notarized universal binary · macOS 15 or newer · MIT licensed

One binary. Two senses.

sense vision for pixels, sense audio for sound. Everything runs locally through Vision, CoreImage, AVFoundation, ScreenCaptureKit, Speech and friends — your files never leave the machine.

On-device, always

Network use is opt-in only, and only for one flag: transcribe --allow-network.

Made for pipes

stdout is data, stderr is status. JSON everywhere, deterministic exit codes, stdin with -.

Trustable builds

Developer ID signed and notarized by Apple, so permission grants survive every upgrade.

Vision

Live Text OCR, face and barcode detection, subject cutouts, smart crops, screen capture — the whole Vision framework, one flag away.

Turn a scanned PDF into Markdown

$ sense vision ocr scan.pdf --md > notes.md

OCR whatever is on screen right now

$ sense vision screenshot | sense vision ocr -

Find faces, get pixel coordinates

$ sense vision detect group.jpg --what faces --json

Read barcodes and QR codes

$ sense vision detect poster.jpg --what barcodes

Blur every face in a photo

$ sense vision filter photo.jpg out.jpg --redact faces --redact-blur

Cut out the person, blur the background

$ sense vision segment photo.jpg --person --blur-bg 25 -o portrait.png

Find near-duplicate photos

$ sense vision dedupe ~/Pictures/Export

Subtitle a screen recording from its pixels

$ sense vision video ocr recording.mov --changes --srt > captions.srt

Audio

Record, transcribe, speak, convert, trim, split — with voice-activity detection and on-device speech recognition. Before your first talk: install the premium voices. They're a free download and sound dramatically better.

Get the premium voices

Opens System Settings where the free premium voices are downloaded — a large quality jump for talk.

$ sense audio voices --install

List installed voices, best first

Add --premium to see just the good ones.

$ sense audio voices

Transcribe a recording, on device

$ sense audio transcribe interview.m4a

Make SRT subtitles from a talk

$ sense audio transcribe talk.m4a --srt > talk.srt

Record until you stop talking

$ sense audio record -o note.wav --vad 2

Have your Mac announce the build

$ sense audio talk "The build finished successfully"

Turn Markdown notes into an audio file

$ sense audio talk -f notes.md -o notes.mp3

Normalize levels before publishing

$ sense audio gain in.wav out.wav --normalize --db -3

The senses compose.

Every command reads stdin and writes clean stdout, so vision and audio chain together like any other Unix tool.

Read a contract out loud

$ sense vision ocr contract.pdf --md | sense audio talk -m

Speak, get text back — no file in between

$ sense audio record -d 5 | sense audio transcribe -

Install in one line.

Homebrew pulls a signed, notarized universal binary for Apple silicon and Intel. Requires macOS 15 (Sequoia) or newer.

Install

$ brew install wkoszek/tap/sense

Grant permissions once

Camera, microphone and screen recording are tied to the code signature, so a grant you give once keeps working across upgrades.

$ sense vision doctor --request

Verify what you installed

Every release is signed with the project's Developer ID and notarized by Apple.

$ codesign -dv --verbose=2 "$(brew --prefix)/bin/sense" 2>&1 | grep TeamIdentifier
TeamIdentifier=QQ5A9Q7C7Z