Ship every lane at once. With your voice.

Keebye is on-device dictation for builders running parallel AI-agent workstreams. Hold a key, say the thing, and the text lands in whichever lane has focus — Claude Code, Cursor, Slack, anywhere you can type.

macOS · early access · audio never leaves your Mac

lane-1 · claude code refactoring the auth module…

lane-2 · cursor tests green, awaiting review

lane-3 · you, by voice "merge lane one, then draft the changelog"

Illustration: three lanes, one voice.

Why this exists

Keebye exists because its founder ran out of hands. Working with high-stakes startups — the kind where a $1M+ investment rides on shipping — meant running several lanes of work in parallel, at home with two kids, one of them under six months old. The only way to multiply output without multiplying hands was voice: dictate into one lane while another lane builds.

Built for the parallel-lane workday

Drive agent lanes by voice

While Claude Code grinds in one terminal, dictate the next prompt into another. Review, redirect, and queue work across lanes without breaking your typing flow in the lane your hands are in.

Reply while you build

Slack threads, emails, PR comments — focus the window, hold the key, and dictate the reply. Then get back to the lane that actually needs your hands.

Hands busy? Still shipping.

Holding a baby, eating lunch, pacing while you think. Hold the hotkey, talk, release — the words land wherever your cursor is.

How it works

  1. 01

    Hold a key, speak

    The default hotkey is Right ⌘ (Fn or Right ⌥ if you prefer). Hold to talk, or tap once to toggle a longer dictation. Esc cancels.

  2. 02

    Transcribed on your Mac

    Speech-to-text runs locally: a fast English-tuned default model, an optional multilingual engine covering 25 languages, or Apple's native engine. Once a model is downloaded, dictation works offline.

  3. 03

    Text lands in the focused app

    Keebye pastes — or synthetically types, which is what you want in terminals, SSH, and tmux — into whatever app has focus. It refuses password fields by design.

  4. 04

    Cleaned up, your way

    Fillers stripped, repeats collapsed, capitalization and punctuation fixed by fast rules — with an optional local-LLM polish pass that also runs entirely on-device.

The honest feature list

  • On-device speech-to-text — audio never leaves your Mac
  • Works offline once the model is downloaded
  • Types into terminals, SSH, and tmux sessions
  • Custom dictionary for your project's jargon
  • Local, text-only history — auto-deleted after 30 days
  • No telemetry, no analytics
  • Lightweight native menu-bar app, built with Tauri
  • Launch at login, stay out of the way

How Keebye compares

Wispr Flow and Superwhisper are both capable dictation tools with real followings. Keebye is built for a narrower job: driving parallel agent lanes with speech that never leaves your machine. Honest breakdowns:

Questions, answered plainly

Does my audio leave my Mac?

No. All speech-to-text engines run on-device. The network is used to download models the first time and to check for app updates — never to send audio.

Does it work offline?

Yes. Once a speech model is downloaded, transcription runs locally without an internet connection.

Which languages does it support?

The default engine is English-tuned. Switch to the multilingual Canary engine for 25 languages — including Spanish, French, German, Portuguese, Romanian, Polish, and Ukrainian — or use Apple's native engine in your system language.

Which apps does it work in?

Anything that accepts text. Keebye inserts by pasting or by synthetic typing into the focused app, with terminal-aware handling for apps like iTerm2 and Warp — and typing-mode insertion works inside tmux and SSH sessions. It deliberately refuses secure password fields.

What platforms are supported?

macOS today. A Windows port exists in the codebase but hasn't shipped a build yet, so we don't advertise it.

How do I get Keebye?

Keebye is in early access — builds are published on GitHub releases. Grab one, dictate for a day, and tell us what broke.

Your hands are the bottleneck. Your voice isn't.

Put a dictation key under your thumb and keep every lane moving.