SkillFoxx

localvoxtral plugin

Local dictation for macOS with Claude Code and Codex plugins: speech becomes prompt text that recognizes your files.

  • 65 ★ on the repository
  • Last commit
  • MIT
Plugin

Extended access

Why this level

  • The app needs macOS permissions for the microphone and for typing into other programs.
  • The plugins pass the app file names, project names and the last session turn, and the install script runs through a curl | bash scheme.
All reasons and checks

t0msilver/localvoxtralAuthor:T0mSIlver

Install

Manual install

curl -fsSL https://raw.githubusercontent.com/T0mSIlver/localvoxtral/main/scripts/install.sh | bash

The README's first option: the script is downloaded and run immediately, read it first. There is also `brew install --cask T0mSIlver/localvoxtral/localvoxtral`.

Access and checks

How we review

Extended accessThe tool runs code or makes network calls and only reads external data.

  • The app needs macOS permissions for the microphone and for typing into other programs.
  • The plugins pass the app file names, project names and the last session turn, and the install script runs through a curl | bash scheme.

This is third-party code. Review the repository files before installing.

What it does

localvoxtral is a macOS menu bar app that turns speech into text in real time, entirely on your Mac. By default it runs the open Voxtral Mini 4B Realtime model from Mistral AI, and on machines with less memory you can take NVIDIA's Nemotron. There is no account, subscription or telemetry. The project is built first for prompting agents. The Claude Code and Codex plugins tell the app what the session is doing, so dictation better recognizes filenames, symbols and the topic of the last turn. A remote-host plugin passes this context through an SSH tunnel. A Claude Code mod puts dictated text into the prompt box and shows the dictation above it. When an agent waits for an answer, the app shows a banner, plays a sound and marks the menu bar icon.

Who it is for. Developers on Apple Silicon Macs who give Claude Code or Codex tasks by voice and do not want audio sent to a cloud.

Good fit when

  • A long prompt to an agent is quicker to dictate than to type.
  • Dictation confuses file and function names and needs the session's context.
  • Claude Code runs on a remote server over SSH while dictation is more convenient on the Mac.
  • Voice recordings must not leave the computer: recognition runs on the device.
  • An agent waits for your decision in another window and a sound should tell you.

Not a fit when

  • An Intel Mac or macOS older than 15: Apple Silicon and macOS 15 or newer are required.
  • You need dictation on Windows or Linux: the app is macOS only.
  • You cannot download 6 GB: on first setup the local engine downloads about 6 GB.

Example request

Turn dictation on with the Fn key and say: "Open settings.py and add a timeout parameter"

Limitations

A Mac with Apple Silicon and macOS 15 or newer are required. On first setup a wizard requests permissions and downloads about 6 GB of models; the engine runs best with 16 GB of memory, and on an 8 GB Mac you turn text polishing off or pick smaller models. The Claude Code plugins declare hooks only, and the text-insertion mod works only on a Claude Code build that loads hooks modules, otherwise it does nothing. The optional project-terms feature launches a separate one-shot Claude Code run, costing roughly 3 to 12 cents per project. Hosted Mistral models and your own servers are allowed, but then audio leaves the Mac.

How to disable. Disable the Claude Code plugin in the app settings or run `claude plugin uninstall localvoxtral@localvoxtral`. Remove the app like any macOS program.

FAQ

How do I install localvoxtral?

With one command: `curl -fsSL https://raw.githubusercontent.com/T0mSIlver/localvoxtral/main/scripts/install.sh | bash`, or via Homebrew `brew install --cask T0mSIlver/localvoxtral/localvoxtral`, or download the DMG from releases. On first launch a wizard downloads the engine.

How do I connect the plugin to Claude Code?

In the app open Settings, the Claude Code section, Plugin and press Install. By hand the commands are `claude plugin marketplace add` with the directory path and `claude plugin install localvoxtral@localvoxtral`. The plugin declares hooks only and spends no tokens.

What leaves the computer?

Nothing by default: recognition runs locally with no account or telemetry. Audio leaves only when you yourself connect Mistral or your own server.

Which agents does dictation work with?

Claude Code, Codex, opencode and Mistral Vibe. Claude Code and Codex have plugins that pass session context, and remote hosts have a separate plugin through an SSH tunnel.

Which recognition models are used?

Voxtral Mini 4B Realtime from Mistral AI by default. On a Mac with less memory you can pick NVIDIA's Nemotron 3.5 ASR Streaming or Nemotron Unified ASR.

Discussion

No questions or reviews yet.

Plugins

Local voice layer for Claude Code and Codex on macOS: when the agent finishes or fails, it says in a sentence or two what happened

CLIExtended access217

Meeting notes on Mac and an MCP for agents: Markdown transcripts, search, recaps and action items, with no bot joining the call

MCP serverExtended access92

A plugin linking Superwhisper and Claude Code: when the agent finishes or waits, a window opens for a voice reply.

PluginExtended access63

A realtime voice layer on top of Claude Code, Codex and OpenCode that keeps talking and listening while background tasks run in parallel

CLIExtended access3.2K

A skill that takes an agent from an idea to a working TypeScript extension for Voz, so Mac dictation becomes commands or notes.

SkillBroad access55

A call-notes app with no meeting bot: transcripts and search are available to Claude Code and Codex read-only through MCP

MCP serverBroad access112

README in short

The project's slogan: talk to your coding agents and keep every word on your Mac. Text appears as you speak rather than after you stop.

README badge

Report a problem