Kekoso is an on-device speech-to-text app for macOS. Every recognition model runs on the machine itself, so there is no account, no API key and nothing uploaded. Four things it does. Press a hotkey and dictate into whatever app your cursor is already in. Drop in an audio or video file, or paste a YouTube link, and get a transcript with timestamps that exports to TXT, SRT or VTT. Record a call and get it split into two tracks, so who said what is a fact rather than a guess. Run a local MCP server that gives Claude Code, Claude Desktop and Codex eight transcription tools. Five models ship with the app: two Parakeet variants for English speed, Whisper small and large-v3-turbo covering up to 99 languages, and SenseVoice for Chinese, Japanese and Korean. Models download once and then work offline. Requires macOS 14.4 or later on Apple silicon. $29 once for three devices, seven-day trial, no card needed. More at https://kekoso.app
Why on-device matters. Cloud transcription meters by the minute and needs an account, so recordings sit on someone else's server and costs grow with use. Kekoso charges once and audio never leaves the machine: a two-hour interview costs the same as a two-minute voice note, nothing. It keeps working without internet, and keeps working if the vendor shuts down.
Who it is for. Developers dictating commit messages without leaving the editor, writers and journalists turning interviews into text, consultants who record calls and need to know which side said what, and anyone who would rather not upload private conversations to a third party. A vocabulary list teaches the app names, product terms and jargon it would otherwise mangle, and those corrections stick across sessions.
