local-stt

What does local-stt do?

local-stt describes itself as "Free, private, offline speech-to-text for macOS and Windows." Hold a push-to-talk key, speak, release, and the text is pasted into the active window. The README says audio stays in memory and never touches the disk or a cloud service.

Three Whisper models are supported, each downloaded once: Whisper Small as the default at 181 MB, Whisper Medium at 514 MB, and Whisper Large v3 at 1.0 GB. The installer is uvx local-stt, which pulls the Python package from the index and runs it, so a one-line command is enough to get started.

Key facts

• Licence: MIT.
• Platforms: macOS 11 Big Sur or later on Apple Silicon (M1+); Windows 10 or 11 on 64-bit CPUs. Linux is not yet supported, per the README.
• Install: uvx local-stt, which needs only the uv installer.
• Speech models: Whisper Small (181 MB, default), Whisper Medium (514 MB), Whisper Large v3 (1.0 GB).
• Minimum RAM: 4 GB for Small, 8 GB for Medium, 16 GB for Large.
• Minimum disk: about 200 MB for the app and the default model.
• CPU floor: an Intel Haswell or AMD Excavator from around 2013, per the README.
• GPU: the built-in GPU on a Mac; any GPU with a Vulkan driver on Windows.
• Audio handling: kept in memory, never written to disk, per the README.
• Pricing: free; the README says "no trial, no paid tier, no usage limits."

What does it replace, and where does it fall short?

local-stt replaces Superwhisper and Wispr Flow for Apple Silicon Mac and 64-bit Windows users who want push-to-talk dictation installed in one line.

Where it falls short: Apple Silicon and Windows only. The README names Linux as not yet supported, so Linux users have to look elsewhere. Intel Macs are not listed. Whisper Large v3 needs 16 GB of RAM and the model is 1 GB on disk, so lower-end laptops are limited to Small or Medium. Post-processing (cleanup prompts, grammar fixes) is not part of local-stt, so the output is raw Whisper text.

How does local-stt run?

Install uv (the Python packaging tool) once with the one-line installer from astral.sh/uv. Then run uvx local-stt in a terminal. uvx creates a throwaway environment, pulls local-stt from PyPI and starts the app. On first launch it asks for Microphone and Accessibility permissions on macOS, or Microphone permission on Windows.

Hold the push-to-talk key, speak, release. The default Whisper Small model handles everyday dictation; switching to Medium or Large v3 is a setting inside the app that downloads the heavier model. Audio is kept in memory and discarded when the model finishes, per the README.

Who is local-stt for?

An Apple Silicon Mac or Windows user who wants a single command to install push-to-talk dictation, with no trial, no account and no subscription. A Linux user should look at Plume or OpenDictate. A user who wants cleanup on top of the raw transcript should look at VoiceInk or OpenWhispr.

What limits does the README state?

• Linux is not yet supported; macOS and Windows only.
• On Mac, Apple Silicon only (M1 or newer); Intel Macs are not listed.
• Whisper Large v3 needs 16 GB of RAM; Medium needs 8 GB; Small needs 4 GB.
• Post-processing (cleanup prompts, grammar fixes) is not part of the app.
• The README gives no accuracy figure; each model's accuracy is on its own card.

Questions people ask

Is local-stt open source?

Yes. The LICENSE file at the repository root is MIT, an OSI-approved licence. The Whisper weights are MIT by OpenAI's release, so the whole stack is clear to use, change and ship. The README adds no restriction beyond the MIT grant.

Does local-stt work offline?

Yes. After uvx local-stt and the one-time model download, the push-to-talk key works with no internet. The README says audio is kept in memory and never written to disk, and no cloud service is called.

Can local-stt be used commercially?

Yes. MIT allows commercial use of the code without a fee. Whisper's weights are MIT by OpenAI, so no separate commercial licence is needed for the speech model. A commercial user who switches to a non-Whisper model should read that model's card.

How does local-stt compare with Superwhisper?

local-stt is free, open source and installed with one command, and it runs on Apple Silicon Macs and 64-bit Windows. Superwhisper is a paid Mac app with its own cleanup prompts and language support. local-stt does the transcription and nothing else; cleanup is a separate step, which keeps the install small. The open-source alternatives to Superwhisper page sets local-stt beside Handy, Dicty and VoiceInk.

Why does local-stt install with uvx?

The README's own answer is that uvx downloads and runs a Python package in one step, with no manual pip or virtual environment. The maintainer chose this to make the install a single line that works on a fresh Mac or Windows PC with only uv installed first.

Sources

• local-stt README and LICENSE: github.com/mariadb-RupeshBiswas/local-stt, read 2026-11-09.