Arietta Voice: open-source local-first framework for building wake-word-driven voice assistants on Apple Silicon Macs

Arietta Voice combines local speech-to-text (Moonshine), text-to-speech (Kokoro), turn detection (Silero VAD and Smart Turn), tool routing, markdown knowledge retrieval, and a web admin console into a customizable assistant runtime designed for always-on Mac mini deployments.

The framework is designed to be customized: users configure the assistant persona in SOUL.md, add knowledge articles, set the wake word, and add deterministic or model-called tools. The default persona is named Bridget; it supports local chat with Gemma on MLX or optional AWS Bedrock chat integration.

The admin console (accessible at http://127.0.0.1:8765/admin) provides observability and editing: health status, logs, diagnostics, chat testing, knowledge editing, tool validation, and tool source editing. The voice runtime and web console can run as separate processes or be managed together via the API.