OpenAI launches Dots, an always-on rival to Meta's Muse

OpenAI launched Dots on September 29, 2026: an always-on AI agent built to rival Meta's Muse, which Sam Altman has said "seems like a nice product." A Dot runs in the background on OpenAI's own cloud computers, watching projects and acting on a user's behalf without needing a chat window open.

TechCrunch reports that OpenAI describes Dots as "remarkably capable, always-on agents built to handle everything." Each Dot gets its own avatar and a name the user picks. Tech AI Wire has covered Meta's Muse running into access-control problems on the Mac, including a Messages sync it made without permission and file access on first launch. Dots now enters that same always-on agent market.

How a Dot works

A Dot is powered by GPT-6 Astra and runs on a dedicated cloud machine rather than the user's own device, according to The Decoder. It can browse the web, analyze data, write code, and draft documents on its own. Users reach it by chatting or calling through ChatGPT, Slack, or Microsoft Teams, and it keeps the same context across all three.

Plugins let one Dot connect to more than 4,000 apps and work on several tasks at once. OpenAI says it is already working with Microsoft to plug Dots into Agent 365, Microsoft's security controls for AI agents.

Built-in limits, for now

The Decoder reports that a Dot defaults to a read-only "proactive research" mode when the user is not actively working with it, and only takes real action after a check. Users can set custom approval rules, and OpenAI runs automated checks before anything sensitive happens.

At launch, each user gets one Dot. The Decoder reports OpenAI plans specialized organizational Dots later, with fixed roles and their own credentials, for company use. Access starts with ChatGPT Pro and Business Premium users; The Decoder reports that Pro subscribers in Europe face regional restrictions that other markets do not.

Catching up to Muse

Meta's Muse launched about three weeks before Dots, and The Decoder reports it added 500,000 users in its first week. Semafor puts its download count above 3 million. Semafor also notes two problems Muse has already run into: Amazon blocked it from shopping on its site, and Meta's own agent disclosed a user's private address without consent.

OpenAI is betting its larger existing ChatGPT user base gives Dots a faster path to scale than Muse had. "Over time, we envision teams of Dots working together on your behalf," the company said, according to TechCrunch.

What this means for developers

Dots is not a developer product today; it ships as a ChatGPT feature, not an API. If you build with OpenAI's models, the news to track is Agent 365. OpenAI's plan to plug Dots into Microsoft's agent security controls suggests a template worth copying: named identities, scoped credentials, and an approval step before sensitive actions.

Muse's stumbles are the more immediate lesson. Both Amazon's block and Meta's address disclosure came from an agent acting with too much reach and too little oversight. Before you give your own agent standing permissions, decide what it should never do without asking first. That is the same read-only-by-default pattern Dots ships with.

If your product competes in this space, watch the organizational-Dots plan. Fixed roles and dedicated credentials per task are a real design pattern for multi-agent systems, not just a feature name.