OpenAI Unveils Dots, an Always-On AI Assistant Built to Challenge Meta’s Muse

Updated September 29, 2026 5:25 pm

OpenAI has introduced Dots, a new always-on AI assistant designed to work across connected apps, browse the web through a cloud computer, and carry out multi-step tasks in the background. The launch, announced during the company’s DevDay keynote on Tuesday, is OpenAI’s clearest answer yet to Meta’s rapidly rising Muse assistant and signals a new phase in the race to build persistent, agentic AI.

OpenAI says Dots can handle a wide range of tasks with minimal supervision, learning user preferences over time and acting across more than 4,000 supported apps through its own cloud-based environment. The rollout begins immediately for ChatGPT Pro and Business Premium subscribers, with a longer-term plan to let users run multiple agents and, eventually, specialized workplace versions.

What OpenAI announced

OpenAI positioned Dots as a background assistant that stays active, remembers context, and can carry out work while the user does something else. The company says the assistant is powered by GPT-6 Astra and is intended to operate more like a persistent collaborator than a single-purpose chatbot.

Unlike a simple conversational interface, Dots can access a cloud computer, open a browser, interact with third-party services, and continue working across sessions and devices. OpenAI says the assistant can also pull in context from ChatGPT interactions on the web, desktop, and mobile, and can communicate through tools such as Microsoft Teams and Slack.

OpenAI describes Dots as assistants that can “do nearly anything” by combining background work, connected apps, and a cloud computer.

The company also said a text-message-style interface will be available, echoing the conversational format that has helped make Muse stand out. A voice option is already available through ChatGPT on desktop, mobile, and web, and text-message support is expected soon.

How Dots works

Dots are built around persistent task execution. Once a user assigns work, the assistant can send progress updates, ask clarifying questions, and continue operating without requiring the user to remain in the chat thread. OpenAI says the assistant is meant to learn from each interaction, adapting to a person’s preferences and anticipating future needs.

That makes Dots closer to an autonomous helper than a reactive chatbot. In OpenAI’s framing, the assistant is not only meant to answer questions but also to complete projects end to end, drawing on app integrations and a cloud environment to move between steps.

Examples OpenAI highlighted

OpenAI offered a few concrete use cases to show how the system could work in practice:

• A software developer could ask a Dot to review user feedback, make changes, test updates, and return a video showing what changed.
• A creator could have a Dot process an interview transcript, identify the strongest clip-worthy moments, draft show notes, and write accompanying social media posts.
• Teams could collaborate inside ChatGPT Space, a new shared workspace where people and their Dot can work together on projects.

The assistant also exposes its cloud computer so users can monitor what it is doing in real time. OpenAI says that visibility is intended to make the assistant more trustworthy and easier to manage than a black-box agent that works entirely out of sight.

Why OpenAI is moving now

The launch comes only weeks after Meta introduced Muse, a highly customizable AI assistant that can browse the web, make purchases, generate documents, and monitor goals. Meta’s push appears to have accelerated competition in one of the most strategically important corners of consumer AI: assistants that do work rather than merely talk about it.

OpenAI’s new product is significant because it shows the company is no longer treating agentic AI as a side feature. Instead, it is building a product class around it, one that combines memory, action, and cross-app continuity. In practical terms, that puts OpenAI in direct competition with companies racing to make AI more personal, more useful, and more difficult to replace.

The timing also reflects a broader industry shift. After years of chatbot launches centered on drafting text or answering questions, the market is increasingly moving toward systems that can execute tasks, connect to software tools, and function more like digital coworkers.

How Dots compares with Meta’s Muse

Dots and Muse share a similar pitch, but OpenAI is emphasizing a few differentiators: cloud computing access, built-in safety rules, collaboration features, and a more explicit workplace angle. Both products aim to keep working after the user leaves the chat, but OpenAI is leaning into structured task management and organization-wide deployment.

The comparison matters because the assistant market is no longer just about benchmarks or model quality. It is now about trust, workflow fit, and whether a product can operate independently without creating new risks for users or businesses.

What safety controls does OpenAI say it built in?

OpenAI says Dots includes built-in rules that govern when the assistant can act on its own. Users can also set custom rules to limit certain actions or force the assistant to ask permission before proceeding.

The company says it has added an auto-review system that checks whether a task complies with the user’s rules and OpenAI’s safety requirements. That review layer is meant to decide when Dots should continue autonomously and when it should hand control back to the user, such as when a password change is required.

OpenAI says the assistant’s review system is designed to verify whether actions comply with custom instructions and safety rules before it proceeds.

This is an important detail. The more capable an agent becomes, the more dangerous it can be if it can act without oversight. Tasks that involve money, private data, account access, or external communications can create real-world consequences if handled incorrectly. OpenAI appears to be signaling that safety is not an add-on but part of the product architecture.

Why safety is becoming the defining issue for agentic AI

Agentic AI is attractive because it saves time. But the same autonomy that makes it useful also makes it more sensitive. If an assistant can browse, message, purchase, or make decisions on behalf of a user, then mistakes can quickly become privacy violations, financial errors, or reputational problems.

That concern is not theoretical. Meta’s Muse has already faced scrutiny after a user said the assistant disclosed their address to a Facebook Marketplace buyer without permission. Incidents like that are shaping how both developers and customers view the next generation of AI products.

OpenAI’s response is to make permissions explicit and keep a visible task trail. The company’s pitch is that users should know what the assistant is doing, be able to set boundaries, and intervene when a task crosses into more sensitive territory.

How many Dots can users run right now?

For now, users can create only one Dot, but OpenAI says that will change. The company plans to support multiple agents in the future, which would allow individuals or teams to assign different assistants to different roles or workflows.

OpenAI also says the output of each Dot can be scaled over time by increasing its speed or the amount of work it can complete each month. That suggests a product roadmap aimed at both casual users and enterprise customers who may want higher-volume automation.

In addition, OpenAI is testing “specialist” Dots for businesses. These versions would be tailored to specific functions inside organizations, hinting at use cases in customer support, operations, marketing, software development, or internal research.

Who gets access first?

OpenAI is starting the rollout today for ChatGPT Pro and Business Premium users. That makes the product immediately relevant to power users and companies rather than the broader consumer base.

The decision also reflects how OpenAI often introduces its most advanced features: first to premium subscribers, then to wider audiences after the company has gathered feedback and refined the product. By limiting initial access, OpenAI can observe real-world usage patterns, identify safety issues, and improve reliability before a broader launch.

Early access details

• Available now to ChatGPT Pro and Business Premium users
• One Dot per user at launch
• Conversations with Dots do not count toward ChatGPT usage limits
• Specialized workplace Dots are in testing
• Text-message support is planned for a future update

What does ChatGPT Space add?

ChatGPT Space is OpenAI’s new collaborative workspace for working with a Dot alongside colleagues. The feature appears designed to make Dots useful inside teams rather than only for individual productivity.

That matters because many of the most valuable AI use cases are shared ones. If a team can hand a Dot a project, review its progress, and iterate together in one place, the assistant becomes less like a novelty and more like part of the workflow infrastructure.

OpenAI’s decision to bundle collaboration into the product hints at a future where agents are embedded in team processes, not just personal chats. That could make Dots particularly attractive for startups, product teams, and media organizations looking for automation that still stays visible to humans.

Why this launch matters beyond OpenAI

Dots is more than a new feature. It is another sign that the AI industry is moving from isolated model demos toward persistent systems that can touch real software, real data, and real decisions.

For consumers, that could mean assistants that help with scheduling, communication, research, and content creation without constant prompting. For businesses, it could mean a new layer of software that sits between people and the tools they already use, handling repetitive work and assembling outputs faster than a human team could.

At the same time, the rise of always-on agents will likely intensify questions about consent, data access, accuracy, and accountability. If these systems can act across dozens or hundreds of apps, the burden on developers to define guardrails will grow quickly.

OpenAI is clearly betting that users want more than answers. The company is betting that they want AI that can work, remember, coordinate, and finish tasks with minimal supervision. Dots is its latest attempt to turn that idea into a product people will trust enough to let run in the background.

Timeline of the Dots rollout

As the rollout begins, the bigger question is not whether AI assistants can talk convincingly. It is whether they can reliably do useful work without causing harm. OpenAI is now making that argument in public, with Dots as the test case.