dot Can Now Connect to Multiple Computers, but Tasks Won't Follow You When You Switch Machines

OpenAI has added multi-computer connections to dot: the same dot can connect to several of your computers, and you specify which one each task uses.

dot is an always-on AI assistant OpenAI announced at DevDay at the end of September. It lives in the cloud and has its own cloud computer and browser. It keeps working even when your computer is off, and you can message it from ChatGPT, Slack, and Teams. Previously it could connect to only one of your own computers; now it can connect to multiple, so you can assign work to a dedicated machine without tying up the one you're using.

What it can do

The OpenAI developer account gave three examples:

• Building an iOS app: have dot test a change in Xcode on a second Mac and send you a screenshot of the result.
• Reproducing a bug: have dot run the same test on your laptop and on a machine with an older SDK, compare the results, and leave your everyday environment untouched.
• Testing the UI: have dot click through a workflow on a second computer and note which step goes wrong. Your screen and mouse remain yours, so you can fix bugs at the same time.

On a connected computer, dot creates a new local Codex or ChatGPT Work task, and can also continue an existing Codex task on that computer. Tasks are managed centrally in the cloud and use the files and tools on that computer.

Each thread uses only one computer at a time. If you want several computers to work independently, have dot open one thread per computer. A task that has already started will not move over when you change the selected computer.

Who can use it and how to connect

dot is gradually rolling out to ChatGPT Pro tiers and Business Premium users. Pro users must be at least 18 years old; it is temporarily unavailable in the European Economic Area, the UK, and Switzerland. Enterprise, Edu, and Healthcare currently require an administrator to enable it, and from November 18 it will be on by default unless the organization chooses to turn it off.

Chatting with dot does not count against ChatGPT usage, while tasks it assigns to Codex and Work count against those products' usage as usual.

To connect a computer: open the ChatGPT app on the computer you want to connect, go to dot's profile page, find this computer under Computers, click Allow, then click Allow access to confirm. To connect a second computer, repeat the process on that computer.

Once connected, the connection remains active, but when dot uses it, that computer must be online and the ChatGPT app must be open. Showing Offline just means it is temporarily unavailable; to disconnect completely, click Revoke access. This authorization is separate from connecting a computer to Codex and enabling Work Sync; enabling just those two does not let dot access your computer.

Compared with Claude Duet and Grok Bot

Anthropic's counterpart to dot is Claude Duet, released on October 21 as a research preview, with US Pro, Max, and Ultra users trying it in a queue. Duet also has its own cloud virtual machine and browser, so it can keep working while your laptop and phone are off, relying on more than 6,000 connectors and plugins.

Duet's official introduction does not mention connecting to your own computer; dot could already connect to one before, and now it can connect to multiple. The official materials also say Duet currently does not let you watch its browser; dot lets you open its cloud computer to watch, and click Take over to take control yourself.

To let Claude use things on your computer, you currently have to go through Claude Code: its project feature lets a task run in a folder on your computer via remote control, but it is centered on code repositories.

xAI's Grok Bot is most similar to dot when it comes to connecting local computers: all Bots share one cloud computer, and after you grant permission it can also execute commands on your computer. Each connected computer is configured separately, and by default every command requires your approval. It can also route the cloud computer's network traffic through your computer, so websites see your computer's IP, which is more convenient for sites that block cloud servers. When dot encounters such a site, its approach is to open a separate local task on your connected computer; login state from the cloud browser is not carried over.

After multiple machines, the trouble moves too

Multi-computer connections put scheduling front and center. One computer per thread and no task migration mean you have to decide first which machine the work should run on. As the number of machines grows, file versions, environment differences, and authorization scope all become everyday issues.

Some users noted that when two computers have the same file, dot may modify the wrong copy; version drift is the cost here. Others said multi-computer switching is handy, but local tasks still require the app to be open on that computer; if it is shut down, the work cannot proceed. The always-on promise holds only for cloud tasks.

One direct change is that previously people watched one computer while waiting for AI to do work; now AI watches several computers while waiting for people to assign work. It sounds easier, but only if you first think through which machines you hand over, which permissions you withhold, and how tasks should be divided.

References: OpenAI Developers' original tweet, dot computer connection documentation. Material from Baoyu's tweet.