OpenAI Dots Explained: ChatGPT Agents That Keep Working Between Chats
OpenAI announced Dots, a new personal agentic assistant for ChatGPT, at its annual DevDay conference in San Francisco, California. CEO Sam Altman introduced the system onstage. Dots is powered by GPT-6 Astra, according to reporting on the launch The Verge.
A Dot is an always-on agent in ChatGPT that takes on ongoing responsibility and keeps making progress between conversations, as described in OpenAI's support documentation OpenAI Help. The idea is persistence. The agent does not reset when a chat ends. It remembers task state, meaning where it left off, and keeps working in the background.
Each Dots agent has its own cloud computer and learns from feedback over time OpenAI. In practical terms, each agent gets separate and isolated online computing space, plus a feedback loop that adjusts its behavior. For people in software and IT, this is long-lived delegation with its own working context, not a simple question-and-answer exchange.
Altman described the agent as "remarkably capable, always-on" AP. The agents look like colorful, personalizable blobs with eyes. OpenAI has limited Dots to users on its $20/month Pro plan and up. ChatGPT has 1.2 billion users The Verge. Altman said Dots are starting as a premium product because they use a lot of compute. He also said the agent has been safety tested Reuters.
OpenAI is working with Microsoft to integrate Dots into its Agent 365 platform. Specialist Dots trained for accounting and legal analysis are planned to launch soon.
DevDay 2026 included more than 20 announcements, including GPT-6 Astra OpenAI. OpenAI also introduced GPT-6.1 Sol as a major upgrade to GPT-6 Sol. Dots was presented as the user-facing packaging for that model work, described in one launch report as a bubbly agentic avatar TechCrunch.
Meta's Muse offers a free dedicated virtual machine that houses both the agent and the user's associated data.
The broader context here is cost and control around always-on operation. Persistent agents use computing power and infrastructure even when the user is away. Someone pays for that use, and someone sets the trust boundary for data the agent can touch between sessions. OpenAI's current answer is paid access plus per-agent cloud computers. Meta's answer is a free virtual machine on its own system.
In my view, the near-term questions for builders are operational. How well task memory holds up across weeks of occasional feedback. How cleanly Dots connect to company login, policy and audit controls inside Agent 365. How specialist training for accounting and legal work keeps domain accuracy without narrowing the general assistant too far. Those integration details will decide adoption more than avatar design.
Looking ahead, if that groundwork holds, the benefit is clear. A billion-scale chat base already knows how to prompt. The next step is learning how to delegate, check, and correct agents that keep working after the window is closed. That shift, from conversation to ongoing responsibility, is what Dots is testing.
