OpenAI Dots Release

OpenAI has launched Dots, a new class of always-on AI agents powered by GPT-6 Astra. These agents are designed to operate independently on behalf of the user, learning preferences over time and executing tasks 24/7 across a vast ecosystem of integrated applications.

Autonomous Execution and Technical Infrastructure

Dots function as autonomous agents capable of handling complex projects without step-by-step direction. This capability is supported by a specific technical stack:

Dedicated Cloud Computing: Each dot operates on its own cloud computer and browser, ensuring a layer of separation between the agent's workspace and the user's local machine.
App Integration: Dots can connect to over 4,000 apps via a plugin ecosystem, allowing them to interact with various software tools to complete goals.
Local Device Access: While they primarily use their own cloud environment, users can grant a dot permission to connect to and use their laptop for direct collaboration.
Cross-Platform Availability: Dots are accessible via ChatGPT (desktop, web, and mobile), Slack, and Microsoft Teams, maintaining full context across all these channels.

Proactive Research and Behavioral Learning

Dots are designed to be an extension of the user, evolving their utility through interaction and feedback. A key feature is "proactive research," where the dot identifies ways to help in the background when the user is not actively engaged. To ensure safety, proactive research is restricted to read-only tools, meaning the agent cannot send messages, modify content, or control the browser/computer without explicit interaction.

Safety, Privacy, and Control Frameworks

OpenAI has implemented several safeguards to manage the autonomy of Dots:

Access and Permission Management

Sandboxing: The dot's cloud computer is separate from the user's computer unless explicitly connected.
Credential Handling: For website logins, dots use saved passwords without exposing them to the underlying model.
Custom Rules: Users can define specific boundaries, requiring approval for certain actions or blocking them entirely. Built-in safety requirements always apply.

Review and Oversight

Auto-Review: This system checks actions that could affect accounts or share information against the user's instructions and safety requirements to determine if an action requires manual approval.
Activity View: Users can monitor background work and redirect the dot as needed.
Sensitive Tasks: Certain high-risk actions, such as changing passwords, are restricted and must be performed by the user.

Data Privacy

Enterprise Privacy: Content from ChatGPT Business, Enterprise, or Edu workspaces is not used to improve models by default.
Training Restrictions: OpenAI does not train directly on proactive research or the dot's internal notes to itself.

Specialist Dots for Enterprise Organizations

Beyond personal agents, OpenAI is introducing "specialist dots." These are agents with dedicated organizational responsibilities, featuring their own identities, credentials, and access to company systems of record.

Early internal testing at OpenAI has focused on procurement, invoice processing, email marketing, customer support, and commercial contracting. These are being rolled out via focused enterprise pilots, with a planned integration into Microsoft Agent 365 to leverage existing enterprise governance and security controls.

Availability and Pricing

Dots are currently rolling out to Pro, Business Premium, and Enterprise users in eligible markets.

Plan Inclusion: The first dot is included at no extra cost for Pro and Business Premium plans.
Usage Limits: Conversations with dots do not count toward standard ChatGPT usage limits. However, tasks managed in Codex or ChatGPT Work do count toward those limits.
Scaling: Future updates will allow users to add more dots and scale individual dot performance by increasing speed or monthly work capacity.