OpenAI Unveils Dots: Persistent AI Agents That Write Code, Fix Bugs, and Research 24/7

At its developer day, OpenAI introduced Dots — permanent agents built on GPT-6 Astra that operate on OpenAI's own cloud computers with a dedicated browser and access to over 4,000 applications via the OpenAI plugin ecosystem. Unlike typical assistants that wait for the next prompt, a Dot continues working after the user leaves, handling multiple projects simultaneously and sharing context across ChatGPT, Slack, and Microsoft Teams. Users can inspect the cloud server's activity or authorize the Dot to run directly from their laptop.

From Feedback to Pull Request

OpenAI demonstrated a Dot taking a task from problem discovery to code delivery. The agent can:

Monitor customer feedback channels

Identify recurring requests

Scope small improvements and bug fixes

Build and test the changes

Return a completed pull request that includes a video walkthrough of the changes

Developers can review these PRs while focusing on larger features. Internally, when a bug appears in Slack, a Dot begins investigating automatically. Another example showed a Dot converting a new design into a runnable application while the team concentrated on customer feedback. Tasks can be queued for a Dot while it is still busy with other work, without creating new threads or step-by-step guidance. The agent's capabilities extend beyond coding: it re-runs scientific analyses as new data arrives, updates sales proposals when customer needs shift, and transforms interview transcripts into video clips, show notes, and social-media posts.

Read-Only Proactive Research

To limit the permissions of an always-on agent, OpenAI separates finding tasks from executing tasks. When not actively directed, a Dot performs proactive search — browsing connected applications to spot opportunities to help — but this access is read-only : it cannot send messages, modify application content, or control the browser or computer.

Any action that could affect the user's account or shared data passes through an auto-review step that checks the action against the user's instructions, OpenAI's safety requirements, and any custom rules the user has set. Rules can allow specific actions to run automatically, require user approval, or block them entirely. Sensitive operations such as changing passwords always remain under direct user control.

OpenAI states that Dots include defenses against malicious instructions — critical for an agent that reads thousands of networked apps — and a monitoring system that can pause or stop the agent when a security issue is detected. All actions, including background work, are visible in an activity view where the user can redirect the agent. OpenAI cautions that Dots can still make mistakes and recommends reviewing the consequences of their actions before proceeding.

Isolation and Data Handling

The user's computer stays isolated from the Dot's environment unless the user explicitly connects it. On supported sites, the Dot can log in using saved credentials without exposing those credentials to the model. By default, OpenAI does not use content from Business, Enterprise, or Education workspaces to improve its models. Personal-plan users control whether Dot conversations and work are used for training; proactive research data and the Dot's own notes are not directly used for training, though they may be utilized if they help complete eligible conversations or tasks, depending on the user's settings.

Specialist Dots and Agent 365

OpenAI also previewed Specialist Dots for organizations. Each specialist Dot receives its own identity, credentials, IT-provided hardware, and access to systems of record, giving it a defined workflow rather than assisting a single employee. Internal pilots have covered procurement, invoice processing, email marketing, customer support, and commercial contracts. External deployment will begin with enterprise pilots where OpenAI engineers collaborate with customers to define each Dot's responsibilities, tool access, and review checkpoints.

Through a partnership with Microsoft, Specialist Dots integrate with the governance and security controls of Agent 365 , allowing administrators to manage these agents with existing tools.

Pricing and Availability

ChatGPT Pro and Business Premium subscribers get one Dot at no extra cost in eligible app markets.

Enterprise customers (including education and healthcare) can enable the beta via workspace admin.

Setup is done in the ChatGPT desktop app or desktop browser; mobile access follows. SMS support is planned.

Conversations with a Dot do not count against ChatGPT usage limits, but tasks launched or managed in Codex or ChatGPT Work still consume those products' limits.

Plans include deeper workspaces and extended usage limits for the first month after launch.

OpenAI intends to let customers add more Dots and increase their speed or monthly work capacity.