OpenAI Dots vs Manus Cue: Which Agent Fits Your Work?

OpenAI dots vs Manus Cue is a choice between an ongoing agent inside ChatGPT and a separate app for a team of personal agents. Start with the accounts you need it to use and the decisions you want to keep. A more interesting avatar does not tell you whether an agent can complete your particular job.

This comparison uses public official documentation checked on 30 September 2026. It is a buying and trial guide, not a hands-on speed test. If the dots terminology is new, what OpenAI dots is explains the agent separately from the model behind it.

Cue is a real product, and it is separate from Manus

Manus’s introduction of Cue describes a standalone personal-agent app built on Manus infrastructure. Each agent has an email address, phone number, wallet and computer. Several agents can work together in a group chat. That is the official meaning of “Manus Cue”; it is not a nickname for the ordinary Manus project workspace.

OpenAI’s dots launch announcement describes a primary agent that works with you across ongoing responsibilities. More dots are a future plan. You should therefore avoid comparing Cue’s described agent team with an imaginary team of dots already available to everyone. The products are developing quickly, and a roadmap is not an account entitlement.

If you want to build and edit a project rather than manage personal agents, read Cue vs Manus 2.0. Those are related products with different starting points. Establishing that distinction prevents paying for a workspace when you mainly wanted an assistant, or expecting a personal-agent chat to be an editing studio.

Access and cost come before the feature list

OpenAI’s access guidance says dots rollout includes Pro outside the EEA, Switzerland and UK, Business Premium in supported ChatGPT regions, and an Enterprise beta enabled by an administrator. Rollout is gradual. A qualifying subscription is necessary, but does not prove the feature has reached your account.

The same guidance says the first dot is included in an eligible Pro or Business Premium plan, with an allowance for deeper work. This does not make a paid subscription free or an agent’s workload unlimited. Check the subscription you already have, then the allowance and any separately metered work you plan to delegate.

Cue’s launch announcement describes free early access with an invitation code and limited places. It also says iOS is awaiting App Store review. Treat both statements as launch conditions, not a permanent free plan or proof that a phone download is available today. If your invitation cannot be redeemed, that is an access issue; it tells you nothing about the quality of the agent.

Before changing subscriptions, list the costs outside the agent itself. A service the agent uses can still charge you. A wallet does not supply a free budget. A cloud computer does not make a paid data source free. Separating those costs is more useful than ranking two monthly price labels that include different things.

Whose identity is doing the work?

Cue’s separate agent identities matter when another person needs to recognise the assistant as a participant. They also raise practical questions: who receives replies, who owns the records, and what happens if you stop using the agent? Make those part of your trial. Do not begin by handing over an inbox full of material unrelated to the task.

At launch, OpenAI’s Help guidance says a dot can use a connected personal email account but cannot have its own standalone email address. That is a concrete difference from Cue’s description. It can affect how you organise an enquiry, even when both tools can research the same options. Check the sender and destination before authorising any message.

An AI agent differs from an LLM because the surrounding system determines tools, identity and actions. A model comparison cannot answer those questions. For a personal assistant, an ordinary reply is only one part of the job; follow-up and ownership are often where the useful differences appear.

Computers and connected accounts are separate decisions

OpenAI’s computers and apps documentation distinguishes its cloud browser, optional access to your computer and connected plugins. Cloud browser sessions do not inherit your personal browser’s logins. Local work requires your computer to stay online with the ChatGPT app open.

This distinction matters for a travel task. Public hotel research might need no personal account. Reading a booking confirmation needs access to the relevant source. Changing the booking is another action with another consequence. “Plan my trip” does not tell the agent which of those steps you intended it to take.

Cue’s public announcement establishes its agents’ own computers and communication identities. It does not provide an exhaustive compatibility list for every service you use. Check the actual connection offered in your account. An independent agent computer should not be treated as a promise that every website accepts automated access or that every existing login transfers across.

A worked example: planning a workshop

Suppose you organise a workshop for colleagues. The job includes venue research, a shortlist, a draft invitation and a list of unanswered questions. An ongoing coordinator could keep these items together. A team of agents could divide the research, shortlist and writing. Both approaches still need a single agreed brief.

Use a request like this as a trial specification:

Prepare a workshop shortlist from the venue pages I supply. Separate confirmed facts from unanswered questions. Use the attendee requirements in this document. Draft an invitation, but wait for my decision before contacting venues, reserving space or sending it. Keep the sources beside each recommendation.

With dots, assess whether corrections carry through the continuing responsibility. Change the accessibility requirement and ask what must be reconsidered. A good result identifies affected venues and open questions instead of merely rewriting the introduction. How to use OpenAI dots gives a fuller method for setting up that kind of assignment.

With Cue, assess handoffs between the proposed agents. Does the writer use the researcher’s actual evidence? Does the shortlist preserve the budget and accessibility constraints? Is it clear which agent is waiting for your decision? More participants can help organise work, but they can also pass along an incorrect assumption unless the shared brief is precise.

Keep the trial fictional or use public information first. There is no need to expose colleagues’ personal details to learn whether the agent preserves a requirement. The purpose is to inspect the workflow before deciding which private sources it needs.

Compare controls by the actions you need

OpenAI’s controls documentation explains automatic action review, app permissions and optional custom rules. It also warns that custom rules are instructions the dot can make mistakes following. They do not grant app access or override built-in safety requirements.

For Cue, the launch page describes payments within the budget you set. That is a capability description, not enough information to conclude that its approval policy matches dots. During setup, inspect the rules actually offered for sending, spending and account access. Ask what happens when a task reaches an action outside the original brief.

Write your boundaries as actions: draft an email, prepare a shortlist, ask before booking. “Be sensible” is difficult to evaluate. OpenAI dots permissions explains why a connected account and an instruction to act should be assessed separately. The same distinction is useful when evaluating another agent, even if its settings use different names.

Which one should you try first?

Try dots first if you already have eligible access and want a continuing responsibility organised inside ChatGPT. Try Cue if you have an invitation and specifically want to explore separate agent identities and group handoffs. Those are fit recommendations based on documented design, not claims that either product is more accurate.

Choose neither for a task whose accounts you cannot appropriately share, or whose result you cannot check. You may be better served by a simple draft that you finish yourself. For a wider personal-agent decision, dots vs Meta Muse considers another documented approach without treating all agents as interchangeable.

Coursium teaches practical AI skills on your phone. Its approach to learning is relevant if the difficult part is specifying work and checking results. It does not replace either agent or supply their subscriptions. Make the smallest useful trial, inspect its output, then decide whether the agent deserves a larger responsibility.