Skip to content
ProjectMakers Logo
ProjectMakersSoftware · Products · Automation

Mac app by ProjectMakers

STTBar

Voice to text in any app.

  • macOS 14+
  • Apple silicon
  • Menu bar app

Press a shortcut, speak, press again. The text appears where your cursor is. Speech recognition runs on your Mac or on a server of your choice.

Coming soon to theMac App StoreFree on GitHub

€1.99 on the Mac App Store or free on GitHub

At a glance

  • Platform

    macOS 14 or later, Mac with Apple silicon.

  • What it does

    Turns speech into text, in any app, right where your cursor is.

  • How to use it

    Press a hotkey, speak, press again. Runs in the menu bar.

  • Privacy

    No account, no tracking. Fully offline on your Mac or with servers you choose.

  • Extras

    Cleanup with a language model, local or through OpenRouter, English output, replacements for names and technical terms.

  • Price

    €1.99 on the Mac App Store, free from source on GitHub.

How it works

  1. 01Speak

    Press the hotkey and talk, the way you would explain the task to a colleague. The overlay shows a live waveform.

Spoken: so um i think we should rework the login flow first store the session in local storage no wait in a cookie then fix the redirect after login and after that please update the auth tests

Cleaned up: Rework the login flow: 1. Store the session in a cookie. 2. Fix the redirect after login. 3. Update the auth tests.

Worked example

Type it or say it?

Say you work with a coding agent all day and write it prompts. Set your own numbers and see how much time goes into typing.

100 words
40 words/min
30 prompts
Typing00:00
Speaking with STTBar00:00+00:00+00:00=00:00
SpeakingTranscriptionCleanup

You save

Per prompt
01:41 min
Per work day
50 minutes
Per month
17 hours

Typing: 100 words ÷ 40 words/min = 02:30 min

Speaking: 100 words ÷ 130 words/min = 00:46 min, plus 00:03 for transcription and cleanup = 00:49 min

Assumptions: speaking at 130 words per minute, 1 second of transcription and 2 seconds of cleanup per prompt, 20 work days per month. Actual times depend on hardware, server and model.

Faster, and the prompt gets better

  • More context

    When you talk, it is easier to explain why something should happen, what you already tried and what must not happen. That is exactly what an agent needs.

  • Every prompt gets checked first

    Before the text reaches the agent, it goes through the cleanup: no filler words, self-corrections resolved, steps numbered.

  • A small local model is enough

    The cleanup does not need a large cloud model. STTBar is tested with Qwen3.5 9B in LM Studio. If the model runs on your side, your text stays with you. If you prefer a hosted model, enter OpenRouter or another OpenAI-compatible service with an API key.

Features

What STTBar does

Three hotkeys, three results

The hotkey you press decides what gets inserted.

  • Cleaned-up text: no filler words, clear structure.
  • Raw transcript: exactly what you said.
  • English output: speak German, paste English.
STTBar settings with the hotkeys for cleaned-up text, raw transcript and English output

Transcription on your Mac or on your server

On the device, STTBar uses WhisperKit and suggests a model that fits your Mac’s memory. No internet connection needed.

Or enter the URL of a Whisper server: on a computer in your network, for example with the included Docker Compose file, or on a server on the internet, with an API key if it needs one.

STTBar transcription settings with a local WhisperKit model

Cleanup with a language model of your choice

If you like, the text goes through a language model before it is inserted: locally, for example in LM Studio or Ollama, or hosted through OpenRouter and other OpenAI-compatible services. Filler words disappear and the text gets a clear structure.

If the model is not reachable, you get the raw transcript.

STTBar post-processing settings with LM Studio

Names and jargon, spelled right

Add word replacements for names and technical terms. STTBar applies them to every dictation, including the raw transcript.

STTBar vocabulary with replacements for product names and technical terms

Privacy

You decide where your recordings go

No account, no subscription, no ads. Recordings and texts only leave your Mac if you enter a server or service yourself. API keys stay in your Mac’s keychain.

Read the privacy policy

History on your terms

The history has its own privacy controls. In sensitive mode STTBar never stores a dictation at all.

Two permissions

STTBar needs access to the microphone and to Accessibility. Accessibility is used only to insert the text into the active app.

How to get STTBar

From the Mac App Store, or free from source.

Mac App Store

€1.99

Buy once, no subscription.

Coming soon to theMac App Store

GitHub

Free

Build STTBar from source or download a notarized build from the releases on GitHub.

License PolyForm Shield 1.0.0: free to use, also in companies. Reselling is not allowed.

Open the repository

Questions or found a bug? Tell us on GitHub

STTBar is a product of ProjectMakers. For our clients we build custom software. More about software development