Daisy-Voice-Agent is a macOS desktop voice-first assistant written in TypeScript. It offers speech input, AI conversation, and local system control with on-device configuration for models and TTS.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
Daisy is a lightweight voice-first AI assistant for macOS that runs outside the menu bar as a floating control. It converts natural language into actions or concise answers, with locally stored API keys and settings. It does not require fixed command words and supports searching, app control, reminders, calendar, and other desktop tasks.
How it works
Daisy uses a voice input workflow triggered by holding a right-side modifier key to capture speech, then sends the result for AI processing and TTS playback. It supports adaptive interactions including natural language chat, app/window control, clipboard read/write, and local searches. All models and speech services are configurable locally via settings; keys/tokens are stored on-device.
Getting started
Installation
"下载 Daisy-1.5.1-arm64.dmg，打开后将 Daisy 拖到“应用程序”文件夹。适用于 Apple Silicon（M 系列芯片）Mac。"
首次运行若遇到 macOS 安全提示，请在“应用程序”中按住 Control 点击 Daisy，选择“打开”，再在确认框中点击“打开”；必要时可前往“系统设置 → 隐私与安全性”选择“仍要打开”。请只从本仓库的 Release 页面获取安装包。
配置服务
首次打开 Daisy，在设置界面填入你自己的服务凭据：
- AI 对话模型：DeepSeek 或兼容 OpenAI 接口的模型配置；
- 语音识别：火山引擎 / 豆包 ASR 的 App ID 与 Access Token；
- 语音播报：默认使用 Microsoft Edge TTS，可在设置中调整；
- 可选唤醒词：使用“嘿 Daisy”时，按提示配置本地
whisper.cpp模型文件。
授权 macOS 权限
按系统提示授予 Daisy 必要权限：麦克风、辅助功能与输入监控、自动化相关权限。
从源码运行
环境要求：macOS、Apple Silicon、Node.js 22 或更新版本，以及 npm。
git clone https://github.com/forestai123456/Daisy-Voice-Agent.git
cd Daisy-Voice-Agent
npm install
cp .env.example daisy.env
# 编辑 daisy.env，填入你自己的密钥；请勿提交该文件
npm run dev
构建 DMG：
npm run dist:mac
产物会生成在 releases/。