OpenAI Unveils 'dots': Always-On GPT-6 Astra Agents with Dedicated Cloud Computers

OpenAI has launched "dots," a significant advancement in AI agent technology, at its recent DevDay event. These are persistent AI agents powered by the company's latest model, GPT-6 Astra. Unlike previous iterations of AI assistants that operate on a request-response cycle, dots are designed to function as always-on, autonomous entities. Each dot is provisioned with its own dedicated cloud computer and a browser, effectively giving it a distinct digital workspace. This architecture ensures that the agent can operate independently, even after a user logs off from their primary device, and maintains its state and ongoing tasks.

The core functionality of dots revolves around pursuing user-defined goals. Users assign a name, a specific objective, and operational boundaries to their dot. The agent then works continuously to achieve this goal, leveraging its access to over 4,000 applications through ChatGPT plugins. This persistent nature allows dots to handle complex, long-term assignments that would be impractical for traditional AI models. While the underlying capabilities for agent-like behavior have been present in earlier technologies such as Codex, OpenAI's innovation with dots lies in the packaging and integration of these elements into a cohesive, persistent agent with a dedicated identity and computing resource. This offers a more streamlined and robust agent experience.

User interaction with dots is designed to be flexible. They can be messaged or called directly within the ChatGPT interface across web, desktop, and mobile platforms. Integrations with popular communication tools like Slack and Microsoft Teams are already in place, with texting functionality planned for the future. A key feature is the "proactive research" mode, where dots can identify opportunities to assist users even when not actively engaged. In this background capacity, connected applications are accessed in a read-only mode to prevent any unintended modifications or actions.

The performance of the underlying GPT-6 Astra model is a critical component. In OpenAI's internal latency simulations, GPT-6 Astra achieved a score of 72.6% on the OSWorld 2.0 benchmark, completing tasks in approximately 40 minutes. This represents a notable improvement over GPT-5.6 Sol, which scored 65.7% in roughly 75 minutes, indicating enhanced efficiency and capability.

OpenAI has also emphasized a robust control and safety layer. This includes built-in rules that govern when a dot can act autonomously and when it should seek user approval. Custom Rules further empower users to define specific actions that can be allowed, blocked, or require explicit permission. Dots are currently available as a managed product, rolling out to ChatGPT Pro and Business Premium users in eligible markets. Enterprise, Education, and Healthcare workspaces will gain access through a beta program, contingent on administrator enablement. Notably, there are no self-hosted or open-weights options currently offered for this product, underscoring OpenAI's controlled approach to its deployment.