Keebye

Keebye

In their words

Keebye is a voice dictation app for developers, writers, and professionals who prefer speaking to typing. Hold a hotkey, speak, and release to insert text into your active application, including terminals, AI coding tools, messaging apps, and document editors.

Speech recognition runs entirely on your computer without uploading audio. After downloading a speech model, transcription works offline. Features include automatic cleanup of fillers and repeated words, a custom dictionary for technical terms, and optional polishing with a local AI model.

Terminal-aware insertion and a typing mode support workflows involving SSH and tmux. Keebye runs on Mac and Windows and offers a 14-day full-featured trial with no credit card required. A paid subscription or lifetime plan is needed afterward.

Fact file

Generator
Price
Free trial
14-day free trial, then $19.99/month or $199.99 Lifetime
Runs on
  • macOS
  • Windows
Sign-up
Account required
Offline
Works without a connection
Built for
developers · writers · parallel AI-agent builders · professionals who prefer speaking to typing
Connects to
  • Claude Code
  • Cursor
  • ChatGPT
  • Slack
  • tmux
Languages
English, Spanish, French, German, Portuguese, Romanian
Trial
14 days, no card
Speech engines
Parakeet, Canary, Apple native
History
Text only, 30-day auto-delete
Model size
~640 MB
Compiled from keebye.com · Nov 2026Spot a mistake? Tell us

What is Keebye?

Speech goes in through the microphone, which is live only while a hotkey is held, and text comes out in whichever app has focus. The default key is Right ⌘ on a Mac or the Right Windows key on a PC, and a single tap starts a longer dictation. Esc cancels. Text lands in applications such as terminals, Claude Code, Cursor, ChatGPT and Slack. Before insertion, fillers, false starts and doubled words are removed by fixed rules. An optional local language model can polish the result, and cleanup can be turned off to keep every word verbatim.

Terminals, SSH and tmux

In a terminal, Keebye pastes with terminal-aware chunking. Turning on synthetic typing lets it work over SSH, inside tmux and on non-QWERTY keyboard layouts. A custom dictionary maps spellings such as gpt4o to GPT-4o and nextjs to Next.js, so product names come out right after they are taught once. On a Mac, the app refuses to type into secure password fields. Windows has no such check yet.

Speech models, languages and what stays on the computer

Parakeet, the default English-tuned engine, takes up about 640 MB at the Int8 setting. The Canary engine covers 25 languages, and on a Mac Apple's native engine can be used in the system language. Once a model is downloaded, transcription works without an internet connection. The network is used for model downloads, update checks and plan confirmation, never to send audio. Local history stores text only and deletes itself after 30 days.

The 14-day trial and paid plans

The trial gives the full app for 14 days from the first desktop sign-in, with no free tier after it. Pro costs $19.99 a month, with a first-month price of $9.99 for new subscribers, and Pro Lifetime is a single $199.99 payment. Sign-in is required. The Windows installer is not yet code-signed, so SmartScreen may warn before it runs.

Similar Products

You may also like