About
Most voice dictation apps today route your audio through cloud servers, run it through third-party APIs, and charge an ongoing monthly subscription.
Dear Machine takes the opposite path. Apple Silicon has enough unified memory and compute to handle speech-to-text and language modeling locally.
Key reasons this model works better:
Real Privacy: Audio inference runs directly on your Mac using Metal. Your voice never leaves your device, so there is no data logging or network exposure. You do not have to rely on trust or privacy policies because the app simply does not send audio over the internet.
Low Latency: Cloud dictation depends on your network speed, server queues, and remote API uptime. Running whisper.cpp locally gives you immediate transcription without network lag, even when you are completely offline.
Structured Output Instead of Raw Transcripts: Pure transcription leaves in stutters, filler words, and awkward phrasing. Running a small local model alongside it cleans up your thoughts into clear prose at the cursor.
One-Time Purchase: Daily utilities do not need to be rental software. You buy the tool once, run it on your hardware, and own it outright.
Cloud AI When You Want It: Local inference is the default, but you can connect your own OpenAI, Claude, or OpenRouter API key when you need a more powerful cloud model.
It is designed for anyone who wants speech to work like a faster keyboard while keeping everything on their own machine. Windows and Linux app is under development.