Keebye

Keebye is push-to-talk voice dictation for macOS. Hold Right Cmd, say the thing, release, and the text lands in whichever app has focus, including your terminal. Speech-to-text runs entirely on your Mac. Tap the key once instead of holding for a longer dictation, and press Esc to cancel. It is built for people running several agent lanes at once, dictating into Claude Code, Cursor, and a terminal instead of typing into each one. What it does - Inserts text into the focused app by paste or by synthetic typing, with terminal-aware chunking that survives SSH, tmux, and non-QWERTY layouts. - Strips fillers, false starts, and doubled words with deterministic rules, plus an optional polish pass that runs a local LLM on-device. Turn cleanup off and you get every word verbatim. - Lets you pick your own push-to-talk key: Right Cmd by default, or Fn, Left Cmd, Left Opt, or Right Opt. - Runs an English-tuned Parakeet engine by default, with the multilingual Canary engine for 25 languages including Spanish, French, German, Portuguese, Romanian, Polish, and Ukrainian, or Apple's native engine in your system language. - Learns your jargon through a custom dictionary, so product names and acronyms are spelled right every time after you teach them once. - Keeps history on disk as text only, on your machine, with automatic deletion after 30 days. - Works offline once a speech model is downloaded. No internet connection is needed to transcribe. - Refuses to type into secure text fields. If the focused field is a password prompt, nothing is inserted. - Sends no audio anywhere. The network is used only to download models, check for updates, and confirm your plan. There is no telemetry and no analytics, and your voice is never used for training. Where it works Anything that accepts text: Terminal, iTerm2, Warp, Claude Code, Cursor, ChatGPT, Slack, and the rest. macOS only today. A Windows port exists in the codebase but no build has shipped. Pricing The trial is free for 14 days with no card required, and starts the first time you sign in from the desktop app rather than when you create the account. It includes the full dictation loop: on-device English transcription, the hold-or-tap hotkey, paste insertion with terminal-aware chunking, the custom dictionary, and local text-only history. Pro is 74.99 dollars per month, currently half off 149.99, and keeps everything from the trial past day 14 while adding the extra models: the Canary engine with 25 languages on-device and the on-device local-LLM polish pass. An annual option is available at 600 dollars, and the monthly price is set to double on October 1, 2026, with your rate locked while you stay subscribed. Pro Lifetime is a single payment of 750 dollars, half off 1,500, for lifetime access to every Pro feature. Pro and Lifetime are the same full product; the only difference is how you pay. Subscriptions can be cancelled at any time and run to the end of the billing period. When the trial ends, dictation stops until a plan is active, and since the trial takes no card, doing nothing costs nothing. Who it is for Engineers running parallel agent lanes who would rather speak an instruction than type it into four windows. Anyone who lives in a terminal over SSH or inside tmux, where most dictation tools fail. People who write long prompts and want them cleaned up without a model reinterpreting them. Non-native speakers and multilingual teams who need 25 languages without sending audio to a server. Anyone under an NDA, a compliance regime, or simply unwilling to upload their voice. Why Keebye Cloud dictation tools send your audio to a server and guess at what you meant. Keebye does neither. Transcription is on-device, cleanup is rule-based by default, the optional polish model is local too, and the app deliberately refuses password fields. What it adds over Apple's built-in dictation is the workflow around it: a real push-to-talk key, insertion that works over SSH and tmux, a dictionary for your project's vocabulary, and history that stays on your disk and deletes itself. keebye.com