# Nova Audio Agent

An always-on, voice-controlled desktop agent that manages coding workspaces and background tasks, deciding when a progress update is worth interrupting the user for rather than narrating every step, with support for macOS, Windows and Ubuntu.

Nova Audio Agent is a harness for an always-on, general-purpose voice agent named Nova. It runs long background tasks, such as coding work handed to Codex, while staying responsive to voice commands, and is built around deciding when a progress update is actually worth interrupting the user for rather than reporting everything.

It supports creating and switching workspaces and sessions by voice with user confirmation, asks clarifying questions before handing work to its background executor, and lets the user add requirements or constraints by voice while a task is running. Its architecture separates a realtime front-end model that interacts with the user from the task-executing back end, sharing state through a common memory store; voice, language and text-to-speech components are pluggable, and it supports custom MCP servers for search and retrieval, personal memory, and a companion iPhone client connected over Tailscale.

It is written in TypeScript and licensed under Apache-2.0, and runs on macOS, Windows and Ubuntu 22.04 or later.

Website: https://deepnovacore.github.io/NovaAudioAgent/