MacHabitsOpenAI launched Dots at DevDay as a personal AI agent: a persistent assistant that lives in ChatGPT, learns your context over time, and acts proactively without being asked. It sits in its own section of the ChatGPT app — separate from regular chats — and can take over your browser or a full cloud computer to complete tasks.
On Mac, it works today. The setup takes under five minutes. What it can do is shaped by what Apple allows third-party apps to access — which is less than on Windows, but still useful.
What Dots actually is
A Dot is a named, persistent AI agent with memory. You give it a name, optionally a visual style, and it remembers what you tell it across sessions. Unlike a regular ChatGPT conversation, a Dot doesn't start from scratch each time.
Dots can take action, not just answer questions. They can browse the web, manage tasks in Slack, take over a cloud-hosted virtual machine to complete multi-step work, and — experimentally — call you on the phone to give updates. The underlying model is GPT-4o.
How to set it up on Mac
- Update ChatGPT to the latest version from the Mac App Store, or download it from openai.com if you haven't installed it.
- Open ChatGPT and look for the Dots section in the left sidebar. If you don't see it, check that your subscription is active under Settings.
- Click Create a Dot. Give it a name. Choose a visual style — there are preset looks or you can customise further.
- Tell your Dot what you want it to know about you: your role, the tools you use, recurring tasks. The more context, the more proactive it becomes.
- To let it see your screen, go to System Settings → Privacy & Security → Screen Recording and enable ChatGPT. You'll be prompted the first time Dots tries to read your screen.
- For Slack integration, connect it inside the Dots settings panel. It will ask for your Slack workspace credentials.
That's the working setup. Everything else — deeper file access, calendar, Mail — requires granting additional permissions if ChatGPT requests them.
What works well on macOS
Browser control works cleanly. Your Dot can open Safari or Chrome, navigate to pages, fill forms, and summarise what it finds — all visible on your screen as it happens. This is the most reliable capability on Mac.
Slack integration is solid. If you use Slack for work, connecting your Dot here is the highest-return action. It can catch up on threads, draft replies, and flag things that need your attention.
The cloud computer is fully available. When you give your Dot a task that involves a lot of steps — research, compiling information from multiple sources, working through a document — it can use a virtual machine in OpenAI's cloud. You can watch it work and take over at any point. This doesn't depend on macOS permissions because it happens in the cloud.
Memory and proactive nudges work regardless of platform. Your Dot will message you when it notices something relevant based on what you've told it.
Where macOS limits it
Apple restricts what third-party apps can do in the background. On Windows, Dots gets deeper access to the file system and other apps through Microsoft Graph and native agent APIs. On Mac, it works through the Accessibility API and Screen Recording — which are opt-in and more limited.
- It cannot read your Mail or Calendar without explicit system permission, and even with permission the access is narrower than on Windows.
- It cannot act on other apps silently in the background. Every action in a foreground app requires the app to be visible.
- File system access requires you to manually grant folder permissions when prompted.
These are Apple's constraints, not OpenAI's. The Windows version of Dots has materially more access to system data. If deep system integration is the main reason you want an AI agent, macOS is the more limited platform for this use case today.
Dots vs Apple Intelligence
They are not competing for the same role. Apple Intelligence is part of the OS — it has no interface of its own and works invisibly inside apps you already use. Dots is a separate app you open and interact with intentionally.
| Dots | Apple Intelligence | |
|---|---|---|
| Model quality | GPT-4o | Apple's on-device model + cloud for complex tasks |
| System integration | Limited by macOS permissions | Deep — it's part of the OS |
| Memory across sessions | Yes, persistent | Limited |
| Proactive actions | Yes | Partial, via Siri suggestions |
| Privacy | OpenAI processes your data | On-device by default |
| Cost | ChatGPT Plus required | Included with macOS 27 |
The practical answer: use Apple Intelligence for quick system actions and writing assistance inside Apple apps. Use Dots for heavier tasks that need a capable model, memory, and the ability to take action across the web and Slack.
Verdict
Dots on Mac is a genuinely useful tool, not a demo. Browser control, Slack integration, and the cloud computer work well within what Apple allows. The gap versus Windows is real but narrow for most practical workflows.
- Set it up, give it your work context, and connect Slack. That alone covers the majority of useful cases.
- Grant Screen Recording permission so it can see what you're working on.
- Keep Apple Intelligence active for system-level tasks — the two tools complement each other rather than overlap.