OpenAI Dots for ecommerce: what teams should know

By Nexscope Team · Published September 30, 2026

OpenAI Dots are always-on agents in ChatGPT that can continue working between conversations, use a cloud computer, and act across connected apps. For an ecommerce team, the near-term opportunity is not an unsupervised store manager. It is a bounded recurring workflow—such as checking a defined set of competitor pages or preparing a daily exception report—with evidence, permissions, and human review designed before the first run.

Key takeaways: OpenAI launched Dots on September 29, 2026. Dots are powered by GPT-6 Astra and can perform scheduled or proactive work, but OpenAI also warns that they can make mistakes. Start with read-only access, define an approval matrix, record sources and timestamps, and measure reviewed outcomes rather than the number of tasks attempted. We found no public documentation confirming a direct Nexscope-to-Dots integration.

What are OpenAI Dots?

An OpenAI Dot is a persistent agent in ChatGPT that can take responsibility for an ongoing goal instead of waiting for a new prompt each time. OpenAI’s getting-started documentation says a dot is powered by GPT-6 Astra, has its own cloud computer, works across apps the user chooses to connect, remembers context, and returns to the user for decisions that require judgment.

That makes a dot different from a normal chat in four practical ways:

OpenAI’s GPT-6 Astra system card adds that Dots can handle recurring work, communicate through supported channels, and delegate parts of work to subagents. The same system card treats proactive, long-running work as a distinct safety problem. “Always on” therefore describes persistence, not guaranteed accuracy or unlimited authority.

Who can use OpenAI Dots, and what do they cost?

Dots are in a gradual rollout, and OpenAI had not published separate long-term Dots usage terms when this article was reviewed. On September 30, OpenAI listed the following initial access:

• Pro: rolling out on web, desktop, and supported mobile experiences, excluding the European Economic Area, Switzerland, and the United Kingdom at launch.
• Business Premium: available across supported ChatGPT regions.
• Enterprise: available as a beta after a workspace administrator enables it; access is initially off by default.

OpenAI said eligible users’ Dots usage would not count toward plan allowances for the first month and that later usage terms would be shared afterward. That is a temporary launch condition, not a permanent free-usage promise. Check the live help page and account interface before budgeting a production workflow.

Users create a dot on desktop web or in the ChatGPT desktop app. Mobile access is part of the rollout, but OpenAI says a new dot cannot currently be created on mobile and Dots are not supported on mobile web. Account, plan, market, and rollout timing may all affect availability.

Which ecommerce tasks are suitable for a first Dot pilot?

Choose a repetitive, reversible, evidence-rich task whose output can be reviewed before it changes a store or contacts a customer. A first pilot should prove that the workflow is dependable; it should not maximize the number of connected systems.

These are pilot designs, not confirmed ecommerce templates supplied by OpenAI. Whether a dot can complete a particular step depends on the connected apps, browser access, permissions, site controls, and the current product rollout. A task that works in a demonstration may fail on a login challenge, a changed page, an unsupported integration, or ambiguous data.

For a broader view of the category, see the AI coworker workflow guide and the agentic commerce overview.

How should an ecommerce team design the first workflow?

Give the dot one decision boundary, one evidence format, and one escalation rule. A useful first specification can fit on one page:

1. Define the goal. For example: “At 9 a.m. each weekday, check these 20 public product URLs and report material price or availability changes.”
2. Name the allowed sources. List the exact apps, folders, dashboards, or URLs. Do not grant broad account access because it may be useful later.
3. Choose a read-only first run. Let the dot collect and summarize before it can publish, purchase, message, delete, or change settings.
4. Require an evidence row. Save the URL or record ID, observed value, timestamp, and reason the item was flagged. Mark missing evidence as unknown rather than filling the gap.
5. Write the escalation rule. Specify when to stop—for example, a login challenge, conflicting prices, a change above 10%, or any request for payment.
6. Run the same task repeatedly. Compare at least several runs for missed changes, false alarms, stale data, and inconsistent classifications.
7. Expand one permission at a time. A reliable report does not automatically justify write access.

This structure also improves the usefulness of the output for later search and AI citations: each conclusion has a named source, date, scope, and uncertainty instead of an unsupported summary.

What permissions and approval controls matter?

Treat a Dot as a persistent operator with access to the systems you connect, not as a disposable chat window. OpenAI documents four custom-rule behaviors: take an action without asking, act when pre-approved in the prompt, ask before acting, or hand the action back to the user. It also states that a dot can still make mistakes while following those rules.

Use “ask before acting” or “hand off” for actions that can affect money, customers, account security, published claims, or irreversible records. A practical ecommerce approval matrix looks like this:

OpenAI says local-computer access is optional and starts turned off. If enabled, the dot can work with local files and local tasks from its messaging channels. Connected apps can also contribute information to memories proactively. Disconnecting an app does not remove information the dot already obtained; OpenAI says deleting the dot through Reset removes its conversations, saved memories, and scheduled tasks. Those behaviors should be part of a team’s access and retention review.

OpenAI Dots vs Meta Muse: what is actually known?

Dots and Meta Muse belong to the same broad “persistent personal agent” category, but they are not interchangeable. Public launch material for both products describes a dedicated execution environment, ongoing work, connected services, and approvals for sensitive actions. Their plan access, interfaces, geographic rollout, controls, and commerce paths differ.

Read the separate Meta Muse merchant readiness guide for its dated evidence and limitations. A comparison based only on launch videos cannot establish which agent is more accurate for a specific store workflow.

Does OpenAI Dots integrate directly with Nexscope?

We did not find a documented direct Nexscope-to-Dots integration in the public OpenAI materials reviewed on September 30, 2026. OpenAI describes connected apps and plugins for Dots, while Nexscope publishes REST and Model Context Protocol tools for compatible clients. Those facts do not prove that Dots can currently call the Nexscope MCP endpoint.

If a team wants to connect ecommerce research data to an agent, first verify the client supports the required remote protocol, authentication flow, tool schema, and approval controls. Then test a read-only call and retain the returned evidence. The ecommerce MCP server guide explains Nexscope’s current endpoint and safety boundaries, but it should not be read as a Dots integration announcement.

How should a team measure an OpenAI Dots pilot?

Measure reviewed work quality and control, not agent activity. A dot that starts many tasks but returns stale or unauditable results creates more work rather than less.

• Task completion: percentage of scheduled runs that return the required evidence format on time.
• Accuracy: confirmed facts, missed changes, false alarms, and unsupported statements.
• Review burden: minutes of human checking and number of approval interruptions per run.
• Control quality: unauthorized attempts, excessive access requests, and correct handoffs at defined boundaries.
• Business outcome: only after reliability is established, measure time saved, faster issue resolution, qualified traffic, or another task-specific outcome.

Do not attribute sales, rankings, or conversion lift to Dots without a controlled comparison and a defensible attribution path. An agent completing a research step is not evidence that it caused the final purchase.

Frequently asked questions

What are OpenAI Dots?

OpenAI Dots are always-on agents in ChatGPT that can keep progressing on ongoing work between conversations. They use GPT-6 Astra, a cloud computer, memories, and selected connected apps, subject to the user’s availability and controls.

How do I get OpenAI Dots?

Create a dot from ChatGPT on desktop web or the desktop app if the feature has reached your account. As of September 30, 2026, rollout is gradual for Pro, Business Premium, and admin-enabled Enterprise access, with regional differences.

Can a Dot buy products or change store data automatically?

A Dot may be able to take actions through supported connected tools or browser flows, but capability and permission are separate. Require explicit approval for purchasing, pricing, publishing, messaging, credentials, and destructive changes, and verify the actual integration before relying on it.

Are Dots included in the ChatGPT subscription price?

OpenAI initially included access for eligible plans and said the first month of usage would not count toward plan allowances. It had not published the later usage terms on the help page reviewed for this article, so check the current plan and account details.

Do OpenAI Dots integrate directly with Nexscope?

No direct integration was documented in the public sources we reviewed. Nexscope’s MCP and REST interfaces can be used only by clients that support their protocol and authentication requirements; confirm compatibility rather than assuming it.

Sources and scope

• OpenAI Help Center: Getting started with your dot, reviewed September 30, 2026 — product behavior, setup, plans, memories, scheduled tasks, connected apps, and controls.
• OpenAI: GPT-6 Astra System Card, Appendix on Dots, updated September 29, 2026 — model, execution environment, long-running-agent risks, and safeguards.
• OpenAI DevDay 2026, September 29, 2026 — launch context.
• Associated Press: OpenAI announces Dots, September 29, 2026 — independent launch coverage and context.

Scope note: Nexscope did not run a controlled Dot ecommerce workflow for this article. The ecommerce examples are test designs derived from OpenAI’s documented general capabilities, not observed performance claims or a promise of integration.

Last reviewed: 2026-09-30 · Maintained by the official Nexscope team.