Stop clicking. Start asking.
Say what you want out loud. Happy does it on your computer, tells you how it went, and shows every step. Free and open source, for Mac and Windows.
Setup is one window: it finds your model, gets the speech models ready, and you press Start. For macOS 13 or later and Windows 10 or 11. It isn't signed yet, so the first time choose Open Anyway on a Mac, or More info › Run anyway on Windows.
Ready when you are
Pick something you do every day. Happy finishes while the old way is still going.
Six steps you get to skip.
Say “Hey Happy” and keep talking, or hold the right Option key (right Ctrl on Windows). Typing works too.
Happy picks the tools for the job. Click its smile and every step is there, with what went in and what came back.
It answers out loud, and says what it's doing on longer jobs. Say “wait” and it holds; ask something new and it listens.
Brightness, volume and play/pause don't even wait for a model, so they happen the moment you finish speaking.
Happy finds what's set up on your computer, and you can add more in a minute.
Link it once with a QR code, then read, search and send messages and files by voice. Away from your desk? Text Happy's own chat from your phone and it does the job on your computer.
Happy speaks MCP, the standard most AI tools use to reach services. Paste the install line a service gives you, and it's ready. Already use Claude Desktop or Cursor? Your services come with you.
It works pages in your own Chrome, signed in as you, and clicks, types and picks menus in your apps by the words on their buttons.
Ask for something to be built and Happy hands it to Claude Code, Codex or another agent you have, in a Terminal window you can watch.
Choose what Happy thinks with. Each choice shows what it needs and where your requests go.
Tell Happy something once and it keeps it. Give it a name and a voice, and it's yours.
“Hey Happy, what's the weather?”
The name is the wake word, so “Happy” in the middle of a sentence to someone else does nothing. Then pick one of 14 voices and how fast it talks.
Happy is free under the MIT license: the app, its prompts and its setup. Read it, change it, and send improvements back.
Install from source
git clone https://github.com/thepropotato/happy.git
cd happy
./setup.sh
On Windows, run .\setup.ps1 instead of ./setup.sh.
Yes. Happy costs nothing. If you use Claude or an API key, you pay that provider as usual; a model on your own machine is free.
It listens for its name and turns speech into text on your computer, acting only when you address it. Turn the wake word off and it listens only while you hold the key.
It does what you asked, and every step is on screen. When a message could reach the wrong person, like a name that fits two contacts or someone you've never messaged, it checks first.
With a model on your own machine, everything on your computer works offline: apps, files, settings and reminders. Web searches, WhatsApp and email need the internet.
Not unless you turn on “Help make Happy better” under Privacy in the settings. Then each turn sends anonymous counts: which tools ran, how long it took, and whether it worked. Never your words, Happy's replies, files or contacts, and the settings show exactly what was last sent.
Yes. Type in the panel, or write to Happy's own WhatsApp chat from your phone. It answers there, quietly.