A computer-use agent for your Mac that works in the background.
Hold ⌘, say what you want, and Yappy does it in the apps you already have: fills the form, books the table, sends the email, files the downloads. It reads the window as text rather than photographing it, works on a window you are not looking at, and stops at logins and anything irreversible. On the Yappy plan, each step is decided by Jev in a few hundred milliseconds.
Free · Apple Silicon · macOS 13+
The same job application, two computer-use agents
One run per column, 18 September 2026, same Ashby form, same answers, same "do not submit". The competitor works from screenshots in the foreground; Yappy worked from the page in the background. Method, caveats and sources: the full study.
What a computer-use agent is, and where most of them are slow
A computer-use agent is an AI that operates a computer the way a person does: it looks at the screen, decides the next action, clicks or types, and repeats until the task is done. Most of them look by taking a screenshot every step and sending it to a vision model. That is general, and it is slow and expensive: the picture is re-sent on nearly every step, the model has to find the button in the pixels, and the window has to be in front of you the whole time.
Yappy takes the other route. macOS already knows what every control on a window is, through the accessibility tree; Chrome and Safari expose the page itself. Yappy reads that as a text table, so the model gets a list of controls with labels and values instead of a photograph, and acts on them by reference. The window can be minimized or on another Space. Nothing on your screen changes until the task is done.
| Property | Screenshot-based agents | Yappy |
|---|---|---|
| What the model sees each step | An image of the screen | A text table of controls and page content |
| Works on a window in the background | No, it needs the window in front | Yes, minimized or on another Space |
| Leaves your cursor and focus alone | No | Yes |
| Per-step decision on the Yappy plan | A vision model, seconds | Jev, a few hundred milliseconds |
| Trigger | A chat box | Hold ⌘ and speak, or type |
| Stops at logins and irreversible steps | Varies | Yes, and says why |
Your model or ours
Agent tasks can run on a Claude Code or Codex install you already pay for, with Yappy driving the Mac, or on the Yappy plan, which includes inference and the Jev decision lane. Dictation runs on-device with Whisper either way, so audio never leaves the Mac. Cleanup of what you said can run on Apple's on-device model. See pricing and the docs.
Computer use on a Mac, answered
What is a computer-use agent?
A computer-use agent is an AI that operates a computer the way a person does: it reads what is on screen, decides what to do, and clicks, types, scrolls, and navigates to finish a task in ordinary applications. Yappy is a computer-use agent for macOS that you trigger by voice: hold the Command key, say what you want, and it does it in the apps you already have.
Which is the fastest computer-use agent for Mac?
The only honest answer is a measured one. On 18 September 2026 we filled the same job application with Yappy and with a screenshot-based competitor: Yappy finished in 1 minute 54 seconds and about $0.24 against 13 minutes 14 seconds and about $4.24, one run each, same form, same answers. The full method, the raw numbers, and the caveats are published on our research page so the result can be repeated and challenged.
Does Yappy take over my screen?
No. Yappy works on the window in the background, including a minimized window or one on another Space, and leaves your cursor and focus alone. It reads the window through macOS accessibility and, in Chrome and Safari, through the page itself, so it does not need the window in front of you.
Does Yappy send screenshots of my screen to a model?
Not for ordinary tasks. Yappy reads the window as a text table of controls and page content, which is what the model sees. Screen images are a last resort, used only when a window exposes nothing readable. On the Yappy plan, each step's decision goes to Jev, which reads text only.
Which AI does Yappy use?
Your choice. Agent tasks can run on a Claude Code or Codex install you already pay for, or on the Yappy plan, which includes inference and the Jev decision lane. Dictation runs on-device with Whisper regardless, so audio never leaves the Mac.
What does it stop for?
Logins, payments, and anything irreversible. Yappy hands control back at those points and tells you why, instead of guessing. It also stops when a field needs a value it does not have.
What does it run on?
macOS 13 (Ventura) or later on Apple Silicon, an M1 Mac or newer. The download is free; the app needs Microphone and Accessibility permissions, which it requests during onboarding.
Free · Apple Silicon · macOS 13+ · Jev in Yappy