parlour

A voice assistant for your home that runs on a Mac you already own. Say the wake word and ask: it hears you with a local wake word and whisper.cpp, thinks with a local model that has tools, speaks back with a local voice, and only hands a question to a cloud model when the local one says it is out of its depth. It controls Home Assistant over MCP, and reaches anything else that has an MCP server too.

You will need a Mac with Node 22 or later and Homebrew. init fetches the rest (ffmpeg, whisper.cpp, llama.cpp and the models) through brew, including a local model sized to the machine, so there is nothing to install by hand and nothing to remember to start.

npm install -g parlour
parlour init          # dependencies, models, config, secrets, and a service or the app
parlour text          # have a conversation in the terminal, no microphone needed
parlour start         # the real thing

Day to day:

parlour doctor            which of the moving parts is down, and how to fix it
parlour restart           after editing config.json or a secret
parlour stop              until the next login
parlour service status    is it running, and does it start at login
parlour service logs      what it has been saying
parlour config edit       ~/.config/parlour/config.json
parlour secrets status    which secrets are set, never their values
parlour connectors list   the accounts your household has signed in to
parlour models suggest    which local model fits this machine
parlour models fetch      the wake word and whisper models

parlour <command> --help shows each command's flags. PARLOUR_HOME moves the config directory; --config <file> or PARLOUR_CONFIG moves just the config file.

As a library

Every stage is a provider behind a port, and a third-party provider is simply an npm package whose default export is a defineProvider({...}):

import { defineProvider, type TextToSpeech } from "parlour";

export default defineProvider<TextToSpeech>({
  kind: "tts",
  name: "parlour-tts-example",
  description: "An example voice",
  create: (options, context) => ({ warm: async () => {}, render: async (text) => renderSomehow(text) }),
});

Name it in config as "tts": { "provider": "parlour-tts-example" } and Parlour loads it on first use, with no change to Parlour itself. Fakes for every port are exported from parlour/testing so you can test yours without hardware.

Documentation

The rest lives in the repository:

• Architecture
• Writing a provider
• Clients: satellites, phones, hardware, automations
• Home Assistant
• The menu bar app
• Tuning
• Your own wake word

MIT. Copyright (c) 2026 Michael Parker.