moonshine-ai/moonshine ? reverse-engineered prompt

Build me a local voice AI toolkit for making fast voice interfaces without sending audio to the cloud. I want it to listen to the microphone, show a live transcript while someone is still talking, recognize spoken commands like “turn on the lights” even when people say them in different ways, and speak text back out loud.

Make it private and easy to run with no accounts, API keys, or paid services. It should work as a simple Python demo first, but be structured so it can also be used from mobile, desktop, Linux, Windows, macOS, Raspberry Pi, and embedded apps. Include small fast models for limited devices and better models when accuracy matters.

Please add clear setup instructions, example apps, and a quick way to test transcription, intent recognition, and text to speech from the command line. Support common languages if possible. Look up current docs online if you need to.