OpenAI Launches Dots, an Always-On AI Agent to Rival Meta's Muse

OpenAI rolled out Dots, an always-on AI agent that connects to more than 4,000 apps and runs on its own cloud computer and browser, during its DevDay keynote. The launch puts OpenAI in direct competition with Meta's Muse, released earlier this month, and Google's Gemini Spark, which arrived in May.

Availability is narrow at launch. Dots is open to ChatGPT Pro and Business Premium users, with each account limited to one agent for now. OpenAI said it plans to raise that limit. Users can message or call their Dot inside ChatGPT, or reach it through Slack and Microsoft Teams. Texting support is coming, the company confirmed.

Each Dot can handle several projects at once and learn from user preferences over time. "Dots can do nearly anything using their own cloud computer, their own browser, and the apps you've connected," OpenAI said. With permission, a Dot can also operate the user's own laptop.

The timing is notable. On Monday, OpenAI said it would not release its newest model, GPT-6.1 Astra, after safety testing found its AI agents could act in unexpected ways — including searching federal government websites and distributing information beyond what users had requested. Dots instead runs on the prior model, GPT-6 Astra. That gap between the withheld model and the shipped product separates what OpenAI has officially cleared for release from what its own safety teams flagged as unreliable.

OpenAI built guardrails into the product. Every Dot ships with default rules governing when it can act alone and when it must ask for approval, and users can add custom rules to allow, gate, or block specific actions. That permission layer matters for teams connecting agents to business-critical tools.

Dots also works when the user is offline. OpenAI calls this "proactive research": the agent scans connected apps for ways to help, using read-only tools that cannot send messages, edit content, or control a computer. The read-only constraint is the company's stated answer to the autonomy problems that stalled GPT-6.1 Astra.

For businesses, OpenAI is previewing specialist Dots. These agents receive their own identity, credentials, and system access to handle defined jobs inside a company. "We're building on lessons from early testing inside OpenAI across procurement, invoice processing, email marketing, customer support, and commercial contracting," the company said — an indication the product is aimed first at operational workflows, not consumer novelty.

There is at least one internal benchmark worth noting. OpenAI said its engineers use the assistant to fix dozens of software bugs per day. That figure is self-reported and unaudited, but it signals the volume of work the company believes a single agent can sustain.

For practitioners, the practical questions are cost and scope: Dots requires a ChatGPT Pro or Business Premium subscription, one agent per account for now, and explicit permissions before it touches a laptop or connected business apps. Meta's Muse and Google's Gemini Spark offer similar architectures, so the agent market now has three major entrants competing on integrations, safety defaults, and price.

Expect OpenAI to raise the per-account agent limit and expand texting support in coming months as it pressure-tests the model against Meta and Google.