It is 11 PM and your shop is closed. A supplier emails a new price list. A regular customer asks on Slack whether her order has shipped. An invoice you sent three weeks ago is still unpaid. Nobody on your team sees any of it until the morning.
On 29 September 2026, OpenAI introduced a product built for exactly this gap. It is called a dot, and it is the company's version of an always-on AI agent. This guide explains what dots are, how they work, what they cost, and what a Hong Kong business owner should check before handing one any real work.
What Are ChatGPT Dots?
In short: A dot is an always-on AI agent that lives inside ChatGPT. It has its own cloud computer and browser, connects to the apps you approve, remembers your preferences, and keeps working on goals you set even when you are not chatting with it. It asks you before taking sensitive actions.
A normal chatbot is like a very smart receptionist who only speaks when you walk up to the desk. You ask, it answers, and then it waits. When you close the window, the work stops.
A dot is closer to an assistant who has a desk, a computer and a to-do list of their own. According to OpenAI's launch announcement, each dot runs on the GPT-6 Astra model, can connect to more than 4,000 apps through ChatGPT plugins, and can work towards your goals 24 hours a day.
The name is a product brand, but the idea behind it is a category you will hear about more and more: the always-on agent. Meta launched a similar agent called Muse on 8 September 2026, and SpaceXAI released Grok Bot in beta on 11 August, as BGR reported. If you want the background on that wider trend, see our earlier explainer on what Meta Muse is.
How Does an Always-On AI Agent Work?
In short: An always-on agent combines three things a chatbot lacks: its own computer to work on, permission to read and act in your connected apps, and memory that lasts between conversations. Together these let it notice what needs doing, prepare the work, and bring it to you for approval.
It helps to break a dot into four simple parts.
1. Its own computer
Each dot works on a separate cloud computer with its own browser. Your laptop stays separate unless you choose to connect it. You can open the dot's computer at any time and watch what it is doing, a bit like looking over a staff member's shoulder.
2. Connected apps
You decide which apps the dot can reach, such as email, calendar, Slack or Microsoft Teams. OpenAI's help centre guide notes that disconnecting an app later does not delete what the dot already learned from it. To erase that, you must reset the whole dot.
3. Memory
A dot remembers context across every channel. Start a task in ChatGPT, continue it in a Slack thread, and the dot still knows what you discussed. Over time it learns how you like things done.
4. Proactive research
This is the "always-on" part. When you are not talking to it, the dot looks through your connected apps for ways to help. OpenAI says this background mode uses read-only tools, so it cannot send messages, change content or control a computer while it is exploring. It can only prepare suggestions.
When the dot wants to act, a system OpenAI calls auto-review checks the action against your instructions and rules. Some tasks, such as changing a password, always stay with you.
What Can a Dot Do for a Small Business?
In short: A dot is best at ongoing, repetitive responsibilities that cross several apps: chasing unpaid invoices, summarising the week, watching for changes, and drafting replies for your approval. OpenAI's own example is a dot that spotted a forgotten invoice, prepared it, and sent it only after the owner approved.
OpenAI tested dots internally across procurement, invoice processing, email marketing, customer support and contract work. Translated into the daily life of a Hong Kong SME, realistic first jobs look like this:
--- Invoice follow-up: every Monday, list invoices older than 30 days and draft polite reminders for you to approve.
--- Weekly owner briefing: every Friday, summarise sales emails, open customer questions and next week's calendar in one page.
--- Supplier watch: flag when a supplier's emailed price list changes, and highlight the items that went up.
--- Meeting preparation: the evening before a client meeting, gather the last emails, quotes and notes into a short brief.
--- Content drafts: turn this week's customer questions into three draft social posts for you to edit.
Notice the pattern. Each job is bounded (one clear responsibility), recurring (it happens every week) and ends with your approval. That is where an always-on agent saves the most time with the least risk.
What a dot is not built for, at least today, is running your front line alone. At launch it cannot call you, cannot have its own standalone email address, and texting is a limited beta for Pro users in the US only, according to OpenAI's help centre.
How Much Do ChatGPT Dots Cost?
In short: At launch, dots come only with ChatGPT Pro and Business Premium plans in eligible markets. Your first dot is included at no extra cost. Chatting with it does not use your normal limits, but the heavier tasks it runs in Codex or ChatGPT Work do. Extra dots will be sold later.
Here is what is published so far:
--- ChatGPT Pro (US$200 per month tier): includes your first dot. The cheaper US$100 Pro tier does not, as BGR noted.
--- ChatGPT Business Premium: reported at about US$100 per seat per month on annual billing, or US$125 monthly, with a two-seat minimum.
--- Enterprise, Edu and Healthcare: a beta that is off by default until a workspace admin switches it on.
--- Coming later: the option to add more dots, or pay for a faster dot or more monthly work.
For a small business, the cheapest realistic entry point is therefore two Business Premium seats, roughly US$200 to US$250 a month, or about HK$1,560 to HK$1,950 at HK$7.8 to the US dollar.
One practical warning for Hong Kong owners: OpenAI says dots are rolling out in "eligible markets" and that some DevDay features may not reach every account yet. Confirm that your account can actually create a dot before you upgrade a plan for it.
What Do People Get Wrong About Always-On Agents?
In short: The biggest mistake is reading "always-on" as "always allowed". A dot is always available and always observing what you connect, but its background work is read-only, and actions go through rules and approvals. The second mistake is assuming it never errs. OpenAI itself says dots can make mistakes.
Misconception 1: "It will run my business while I sleep."
Not by default. Background work is limited to reading and suggesting. Anything that sends, edits or buys goes through the rules you set: take action without asking, take action if pre-approved, ask first, or hand off to you.
Misconception 2: "The included dot means unlimited work."
The first dot is included, but OpenAI has not published an unlimited allowance. Tasks it runs in Codex or ChatGPT Work count against your plan's usage limits, and launch-month limits are temporarily extended.
Misconception 3: "If I disconnect an app, it forgets everything."
It does not. Information the dot already gathered stays in its memory until you reset the dot, which deletes its conversations, memories and scheduled tasks.
Misconception 4: "It is just a smarter chatbot."
A chatbot answers. An agent takes responsibility for ongoing work. That difference is why the setup questions are about permissions and review, not about prompts.
How Should a Hong Kong SME Prepare for an Always-On Agent?
In short: Start with one small, recurring job, connect only the apps that job needs, write down what the agent may do alone and what needs your approval, and check your customer-data obligations under Hong Kong's privacy law. Review the first few completed tasks line by line before trusting it with more.
Whether you choose a dot or any other always-on agent, the same five-step checklist applies:
--- Step 1, pick one job. Choose a task that repeats weekly and has a clear "done", such as the Monday invoice list.
--- Step 2, connect the minimum. If the job only needs email and your accounting export, do not connect WhatsApp, your drive and your calendar as well.
--- Step 3, write an approval table. Three columns: things it can do alone, things it must ask about, things it must never do. Anything that involves money or customers goes in the second or third column.
--- Step 4, check privacy. Under the Personal Data (Privacy) Ordinance, you remain responsible for customer data an overseas service processes on your behalf. Tell staff what not to paste in.
--- Step 5, review, then widen. Read the first three completed tasks in full. Only add a second job once the first one runs without corrections.
A useful comparison is hiring a temp. You would not give a new temp your bank login and every customer file on day one. You would give them one desk, one task and a supervisor. An always-on agent deserves the same onboarding.
If your team already used Custom GPTs, note that OpenAI is also moving businesses from GPTs to plugins. Our guide on what replaces Custom GPTs explains that change.
Frequently Asked Questions About ChatGPT Dots
Is a dot the same as a ChatGPT chat?
No. A chat stops when you stop typing. A dot keeps making progress between conversations, runs scheduled tasks, and can message you with updates or questions.
Can a dot use my own computer?
Only if you allow it. Access to your local computer starts switched off, and you can revoke it at any time from the ChatGPT desktop app.
Does OpenAI train on my dot's work?
OpenAI says it does not use content from Business, Enterprise or Edu workspaces to train its models by default, and does not train directly on proactive research or a dot's private notes. Personal plans have their own data controls.
Can I pause or delete a dot?
Yes. You can pause it from its profile menu, and resume later. Reset deletes the dot with its conversations, memories and scheduled tasks.
Do I need a technical person to set one up?
Setup is done in the ChatGPT desktop app or desktop browser through guided prompts. The harder part is not technical. It is deciding which job to delegate and which approvals to require.
Conclusion: A Colleague That Never Clocks Out Still Needs a Manager
In short: ChatGPT dots are OpenAI's always-on agents: AI assistants with their own computer, memory and connected apps that keep working between conversations. They are powerful for recurring, bounded jobs, but they only suit businesses that set clear permissions, start small and review the work.
The shift from chatbot to agent is the most important change in business AI this year. Your question is no longer "what should I ask the AI?" but "what responsibility can I safely hand over, and how will I check it?"
Pick one late-night headache in your business this week, such as the unpaid invoices or the Friday summary, and write down exactly what "done" looks like. That single page is the job description any always-on agent will need, whichever product you choose.
And if you would rather not work out the permissions and privacy rules alone, you do not have to. We understand AI. UD stands with you.
Reviewed by the UD AI team.
Before you hand any always-on agent real work, it helps to know which jobs in your business are ready for it. Take the free AI Ready Check, and we will walk you through it step by step, from choosing the first task to setting approvals your team can trust.
