Skillget

larmor

mcp
Security Audit
Fail
Health Warn
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 5 GitHub stars
Code Fail
  • eval() — Dynamic code execution via eval() in engine/larmor_engine.py
  • rm -rf — Recursive force deletion command in web-install.sh
Permissions Pass
  • Permissions — No dangerous permissions requested

No AI report is available for this listing yet.

SUMMARY

Talk to your coding agent, and have it talk back while it works. Runs entirely on your Mac.

README.md

Larmor

Talk to your coding agent, and have it talk back while it works.

Larmor gives any agent that speaks MCP a voice and ears: Claude Code, Claude Desktop, Codex,
Gemini CLI, Antigravity, or anything else that can add an MCP server. You can talk over it and it stops; say
"mm-hmm" and it carries on. What you say while it's busy editing files reaches it mid-task.
Everything runs on your Mac. No audio leaves it, and there are no API keys.

larmor.dev

Install

Apple Silicon Mac, macOS 14 or later, about 6 GB of free disk. The engine uses about 3 GB of
memory while it runs; 16 GB Macs are comfortable, 8 GB works but is tight.

curl -fsSL https://larmor.dev/install.sh | bash

That one-liner downloads this repo as a tarball from larmor.dev and runs install.sh.
If you'd rather read it first, or run it straight from source:

git clone https://github.com/naveenvasou/larmor.git ~/.larmor/app
~/.larmor/app/install.sh

The installer asks for an email (optional, Enter skips it), finds the agents on your Mac and lets
you pick which ones get Larmor, then downloads about 3.7 GB of models with a progress bar. The first time you use it,
macOS asks your terminal for microphone access.

Start talking

Open a new session in your agent:

  • Claude Code: type /larmor
  • Claude Desktop: quit and reopen it (⌘Q), then ask for "voice mode" in a new chat
  • Codex: type $larmor. Run /hooks once and trust Larmor's hooks.
  • Anything else: ask it for "voice mode"

For an agent the installer doesn't know, add the output of ~/.larmor/app/install.sh --print
to its MCP settings.

What the installer changes on your Mac

Where What
~/.larmor/ The app, its Python environment, your email, and logs
~/Library/LaunchAgents/dev.larmor.menubar.plist The menu-bar app, started at login. It runs the speech engine on 127.0.0.1:8160
~/.cache/huggingface/hub/ The voice models (Parakeet, Chatterbox)
~/.local/bin/uv uv, only if you don't have it already
Claude Code claude mcp add larmor -s user, the skill in ~/.claude/skills/larmor, hooks in ~/.claude/settings.json
Claude Desktop mcpServers.larmor in ~/Library/Application Support/Claude/claude_desktop_config.json (backed up once to claude_desktop_config.json.bak-larmor)
Codex A [mcp_servers.larmor] block in ~/.codex/config.toml (backed up once to config.toml.bak-larmor), hooks in ~/.codex/hooks.json, the skill in ~/.agents/skills/larmor
Gemini CLI mcpServers.larmor in ~/.gemini/settings.json, the skill in ~/.agents/skills/larmor
Antigravity mcpServers.larmor in ~/.gemini/config/mcp_config.json, the skill in ~/.gemini/skills/larmor

Only the agents you pick are touched, and only Larmor's own entries in their config files.
A config file that isn't valid JSON is left alone.

The hooks are small Python scripts in hooks/. One keeps a voice turn from ending
without listening, one records which session turned voice mode on, and one hands the agent
anything you said while it was working. They do nothing in sessions where voice mode is off.

What leaves your Mac

  • Your email, once, at install, only if you leave one. It's how you hear about updates while
    Larmor is in beta.
  • Feedback, only if you type it into "Send feedback…" in the menu bar.

Both go to one small endpoint whose entire code is backend/signup.py.
Nothing else is sent: no audio, no transcripts, no usage data. The only other network traffic is
downloading the installer, Python packages and the models.

How it works

  • Ears: Apple's VoiceProcessingIO echo canceller, Silero VAD, Smart Turn v3 for end-of-turn,
    and Parakeet (MLX) for speech-to-text.
  • Voice: Chatterbox Turbo (MLX), played through the same echo canceller, so the mic hears you
    and not the agent.
  • Interruptions: your voice pauses playback at once. Keep talking, or say something real like
    "stop", and it stops. A backchannel like "mm-hmm" resumes the sentence.
  • Talk while it works: what you say mid-task reaches the agent with its next spoken progress
    update, in any agent. In Claude Code and Codex it also arrives after the next tool call, usually
    within seconds.
  • One session at a time: voice mode belongs to the session that turned it on. Your other
    terminals behave normally.
  • Lets go of the mic: voice mode switches itself off after 10 minutes without the agent
    using it, or 20 minutes of nobody speaking, so a forgotten chat can't keep your mic open.

The audio front end is a small Swift program, native/larmor_audio.swift.
The repo ships it prebuilt; to build it yourself:

cd native && cp larmor_audio.swift main.swift && swiftc -O main.swift mic_arbiter.swift -o larmor_audio && rm main.swift

Uninstall

~/.larmor/app/install.sh --uninstall   # stops the app and removes the login item
claude mcp remove larmor -s user       # Claude Code

Then, for the agents you set up, delete the larmor skill folders and Larmor's entries in the
config files listed above. To free the disk space, delete ~/.larmor and the Parakeet and
Chatterbox folders in ~/.cache/huggingface/hub.

Can't see the menu-bar icon? A full menu bar hides new icons. Hold ⌘ and drag some out.

Licence

MIT. The bundled models and the ones downloaded at first launch keep their own
licences; see THIRD_PARTY_NOTICES.md.

Reviews (0)

No results found