13.0.0 — Node 24.15, and quieter voice input

A snapshot as of 13.0.0 (released 2026-11-18). It will go stale as the app moves on — that is expected, and the living reference is the guide each section links to.

Before you upgrade: this release needs Node.js 24.15 or later. Check with node -v. If it prints v22… or a v24 below v24.15.0, install the current Node LTS first (Getting started, step 1). On an older Node, npx mulmoterminal stops with a “Node too old” message and the steps to upgrade.

Upgrade:

npx mulmoterminal@latest

New features

Voice input can skip silence and noise

Voice input sends every stretch the browser hears as loud enough, so background noise reaches the transcriber too. Given silence or noise, Whisper tends to write a stock phrase it saw often in training — “Thank you for watching”, “ご視聴ありがとうございました” — into your prompt (#3296).

You can now pass extra flags to the whisper-server that voice input uses, with the WHISPER_SERVER_EXTRA_ARGS environment variable (#3321). With whisper.cpp’s speech detection on, the parts with no speech are skipped.

To try it (macOS, where voice input runs):

  1. Check that your whisper-server has speech detection — this should print a line:
    whisper-server --help 2>&1 | grep -- --vad
    
  2. Download a Silero speech-detection model from ggml-org/whisper-vad.
  3. Set the variable in the shell that starts MulmoTerminal, or in the .env beside it, and restart MulmoTerminal:
    export WHISPER_SERVER_EXTRA_ARGS='--vad --vad-model /path/to/ggml-silero-v6.2.0.bin'
    
  • A path with spaces: write a JSON array, '["--vad", "--vad-model", "/path with space/x.bin"]'.
  • Without a model, --no-speech-thold 0.4 also helps (the default is 0.6).
  • Flags MulmoTerminal sets itself (--model, --host, --port …) are refused: the server log says WHISPER_SERVER_EXTRA_ARGS ignored, and voice input runs as before.
  • A flag your whisper-server does not accept can stop it from starting. If voice input stops working, remove the variable and restart.

Nothing changes unless you set the variable. More: FAQ — voice input types text when I am not speaking.

What looks different

Nothing on screen. The one visible difference is at start-up: on Node older than 24.15, npx mulmoterminal now stops with the “Node too old” message where it used to start (#3321).

Under the hood

Node.js 24.15 or later is required

MulmoTerminal shares its server core with MulmoClaude, and that core now uses Firebase 13, which needs Node 24.12 or later. MulmoTerminal follows MulmoClaude to 24.15 (#3321). This is why the version is 13.0.0: anyone still on Node 22 has to upgrade Node before upgrading MulmoTerminal. The remote-server Docker example in Run the server on another machine now starts from node:24.

Dependencies

The shared MulmoClaude packages move to their new majors along with that core, and Firebase (used by the phone connection and shared apps) moves to 13.4.0. Nothing needs doing; if Google sign-in for the phone connection misbehaves after the upgrade, please report it. Routine patch and minor updates as well (#3319).

How to tell you have the release: npx mulmoterminal@latest starts on Node 24.15 or later, and with WHISPER_SERVER_EXTRA_ARGS set, the server log shows WHISPER_SERVER_EXTRA_ARGS applied as it starts.


This site uses Just the Docs, a documentation theme for Jekyll.