Can I dictate straight into my editor and terminal?

Yes. Keebye inserts text into whatever has focus, and the terminal path chunks the output so shells and TUIs receive it cleanly. Your editor, a commit message in the terminal, and a browser PR form all behave the same.

Is it safe to use around proprietary code?

That is the point of on-device transcription. Your speech is turned into text locally, with no telemetry and nothing uploaded, so describing an internal system or a private repo out loud never sends a byte off your machine.

Will it mangle identifiers, snake_case, or CLI flags?

Feed the custom dictionary your repo names, package names, and the flags you say often, and they land correctly. Dictation shines for the prose around code — messages, comments, docs — more than for typing symbol-dense code itself.

Does it refuse to type into password prompts?

Yes. Keebye detects secure and password fields and declines to insert there, so a dictation aimed at the wrong window will not spill a token or secret into a credential box.

What model handles non-English or mixed-language work?

Parakeet covers English by default. Opt into Canary and you add 24 more languages, which helps if your comments or team chat are not in English.

Voice Dictation for Developers on macOS

Why developers dictate the prose around the code

Most of a developer's writing is not code. It is the commit message explaining why, the PR description walking a reviewer through the change, the comment that saves the next person an hour, the doc nobody wants to write, and the growing pile of prompts you feed your tools. That is a lot of typing, and none of it is the part you were hired for.

Speak it. A commit body is faster said than typed, a PR description comes out in a couple of spoken paragraphs, and the terminal where your agent lives takes dictated input as readily as a text field. Keebye is not for hammering out symbol-heavy code by voice — it is for the human-language layer wrapped around it, which is where the hours quietly go.

How it works

Focus the field — editor, terminal, review form — hold Right ⌘, and speak. On release the text is inserted where your cursor is. In the terminal, insertion is chunked so shells and full-screen TUIs receive it without garbling, and secure fields are refused outright so a stray dictation never lands in a password prompt.

Transcription is fully local. Parakeet downloads once and then works offline, with no telemetry, which is what makes it usable against proprietary code: you can narrate a private architecture or a client's repo and none of it is uploaded. A local history keeps your recent text for thirty days and then deletes it automatically.

Setup in two minutes

Grant Accessibility and microphone access on first run, choose your hotkey — Right ⌘, Fn, or Right ⌥ — and switch on launch-at-login so it is ready in every session. That is the install.

Then invest a minute in the custom dictionary, which pays off fastest for developers: load your repo names, service names, framework terms, and the CLI flags you dictate constantly, so identifiers transcribe right the first time instead of becoming a find-and-replace chore.

Limits, honestly

Dictation is a poor fit for typing code itself — brackets, operators, and camelCase are slow to speak and easy to garble. Use it for messages, comments, docs, and prompts, and let your editor and AI tools handle the symbols, the way a vibe coder leans on voice for intent and the tool for syntax.

It has no editor or Git integration; it types text and you drive everything else, including in Cursor. Text appears when you release the hotkey rather than streaming as you talk — transcription is batch. And Keebye is macOS only.