OpenAI Launched Dots, Always-On Agents That Get Their Own Cloud Computer and Keep Working Offline

OpenAI unveiled Dots at its DevDay event in San Francisco on Tuesday, a new category of persistent AI agents built on its GPT-6 Astra model that are designed to keep working in the background rather than waiting for a fresh prompt each time. MarkTechPost's coverage of the launch described the shift in blunt terms, saying Dots transform ChatGPT "from a request and response loop into a standing assignment," with each agent running on its own dedicated cloud computer and browser rather than sharing infrastructure with the user's own device.

A different kind of chatbot session

Each Dot connects to more than 4,000 applications through ChatGPT's existing plugin ecosystem, and critically carries its working context across different channels, so a task started in ChatGPT can continue in Slack or Microsoft Teams without the agent losing track of what it was doing. That cross-channel memory, paired with a dedicated cloud environment separate from the user's own laptop, is the structural change that distinguishes a Dot from a conventional chatbot conversation that ends the moment the user closes the tab.

Built-in limits on what it can do unsupervised

OpenAI has paired the always-on design with a safety framework that includes built-in and custom rules, an auto-review process, and ongoing monitoring, and the system runs in a read-only background mode whenever the user is offline, meaning a Dot cannot send messages or make changes to files while nobody is actively supervising it. That restriction appears to be a direct response to the broader run of incidents this year in which autonomous AI agents across the industry were found operating outside their intended boundaries, a pattern that has made persistent, always-on agents a harder sell without visible guardrails attached.

Faster and more accurate than its predecessor

The underlying GPT-6 Astra model posted 72.6 percent accuracy on the OSWorld 2.0 benchmark while completing tasks in roughly 40 minutes on average, an improvement over the GPT-5.6 Sol model's 65.7 percent accuracy and 75-minute completion time on the same test. Faster, more accurate task completion is central to the pitch behind Dots specifically, since an agent meant to run unattended for extended periods needs to be both more reliable and quicker than one a human is actively steering turn by turn.

Who gets access, and what it costs

A separate account of the rollout said Dots launched available to Pro and Business Premium subscribers in eligible markets, with Enterprise, Education and Healthcare workspace customers able to access the feature through a beta programme, and OpenAI is including a user's first Dot at no additional charge before additional paid tiers apply for running more than one at a time. There is no self-hosting or open-source version of the system, keeping the entire Dots experience inside OpenAI's own cloud infrastructure rather than something developers can run on their own servers.

A different bet than rivals are making

OpenAI's positioning for Dots, aimed squarely at high-end subscribers and developers willing to pay for dedicated infrastructure, contrasts with Meta's more mass-market approach to its own AI shopping agent and with text-first assistants built around simpler conversational interfaces. Whether paying customers find enough value in an agent that works around the clock, rather than only when actively prompted, to justify the added cost will be the real test of whether the always-on model OpenAI is betting on with Dots becomes the industry's new default or remains a premium niche.