What is daisy-voice-agent?

Daisy is a voice first AI assistant built for macOS on Apple Silicon. It appears as a small floating ball that lives outside the menu bar and is always ready to listen. You hold down a key, speak a request in natural language, and let go. Daisy turns your speech into either a spoken answer or an action on your computer, without requiring fixed command words. It can hold a natural spoken conversation and read the answer back out loud, open, close, or hide apps, send keyboard shortcuts, type text, adjust volume, control media playback, or lock the screen. It can jump straight to search results inside specific popular sites instead of a generic web search, such as searching a video site or a podcast app directly. It also handles everyday tasks like creating and searching memos, reminders, and calendar events, checking weather, maps, or the time, and reading or writing the clipboard. Within permissions you grant, it can read, create, and organize files, work with documents and PDFs, download public videos or audio, and search for wallpapers. The floating ball shows its current state, such as listening, thinking, speaking, or idle, so you always know what it is doing. To set it up, you download a DMG file from the project's releases page and drag the app into your Applications folder. Because the release is only ad hoc signed and not notarized by Apple, macOS may block it on first launch, requiring you to manually allow it through system settings. After installing, you provide your own credentials for the services it uses: an AI chat model such as DeepSeek or an OpenAI compatible API, a speech recognition service, and a text to speech service, plus an optional local wake word model if you want hands free activation. You also need to grant macOS permissions for microphone access, accessibility, input monitoring, and automation of system apps. The project can also be run from source using Node.js 22 or newer, and is released under the MIT license, though bundled components like Electron, whisper.cpp, and yt-dlp keep their own separate licenses.