Introduction
OpenAI released Dots on September 29, 2026 during DevDay 2026. The product family is collectively named Dots, while each individual persistent agent instance is referred to as a dot. Powered by GPT-6 Astra, every dot comes with an independent cloud computer and browser environment. It can continue task execution across separate chat sessions, invoke backend agents, and pause to request human confirmation when critical judgment points emerge. Dots is being rolled out gradually, and access availability depends on user subscription tiers, geographic regions, and workspace configuration. This article breaks down core capabilities, operating logic, setup workflows, access eligibility and security boundaries of Dots, based on OpenAI DevDay presentation materials and official documentation as of September 30, 2026.
What Exactly is OpenAI Dot?
OpenAI Dot is not a conventional chatbot that only responds to isolated one-off questions. It is defined as an always-on persistent AI agent. The core distinction is task continuity: users assign a long-term objective, specify actions that can run autonomously and items requiring human review. The agent tracks progress continuously and adjusts execution as external conditions shift over time.
A common source of confusion is terminology. Dots refers to the entire product line, while a dot represents a single persistent agent owned by one user. In this context, “Dot” and “OpenAI Dots” both point to this new persistent agent capability, unrelated to Graphviz Dot language, Dart programming language or other similarly named tools.
Traditional chat interactions follow a simple turn-based pattern: users send a query, and the model replies within that session. A dot operates differently. Once a long-running responsibility is delegated to it, the agent will continue advancing the task and send status updates across multiple conversations over hours or days.
How Dot Differs from Standard ChatGPT and Scheduled Automation
The key innovation brought by Dot is transforming one-time prompt interactions into sustained work relationships with memory, cross-session tracking and orchestration of auxiliary subtasks. The table below contrasts its workflow characteristics against ChatGPT and regular scheduled jobs.
| Method | Task Progression Logic | Suitable Scenarios |
|---|---|---|
| Standard ChatGPT | Responds only within the active chat session. Requires new user messages to proceed. | One-off Q&A, essay drafting, static analysis |
| Scheduled Tasks | Runs strictly according to pre-configured time triggers. | Daily summaries, recurring reminders, periodic status checks |
| Dot | Tracks long-term goals independently between chat sessions. It can pause, resume and notify users when human judgment is required. | Project tracking, document maintenance, cross-workstream long-running tasks |
OpenAI documentation explicitly states that a dot does not rely on rigid timetables for every subsequent operation. It autonomously decides when to pause execution and recheck task status. It can also schedule calendar events for workflows that need fixed timing. Closing a chat window or voice conversation will not automatically terminate assigned work.
How Dot Sustains Long-running Work
Four core components enable cross-session persistent execution: continuous memory, dedicated cloud computer, backend agent orchestration and cross-channel communication pipelines. These modules turn discrete chat messages into extendable long-term workflows.
1. Independent Context and Note-taking
A dot retains relevant ChatGPT conversation history, saved user preferences, action decisions and in-progress notes. These notes are not full transcript records, but condensed context that persists across separate conversations.
Users can interact with the identical dot instance through ChatGPT, Slack and Microsoft Teams. Switching communication channels does not spawn a new agent or erase its stored notes. Message visibility remains separated by channel, and cross-channel sharing of private data is still constrained by permission limits.
2. Dedicated Cloud Computer and Browser
The built-in cloud computer enables the agent to research information, process files and run software. Task execution continues even after the user’s local device powers off. When website login is required, users can supply credentials through private login flows or complete authentication in the embedded browser before handing operational control back to the dot.
Users have an optional method to connect a personal local computer, allowing the dot to access local files, source code and desktop applications. According to OpenAI documentation updated on September 2026, only one personal desktop may be linked to a single dot at any time. For local task execution, the connected computer must stay online, and the ChatGPT desktop application must remain active.
3. Backend Agent Scheduling
Dot can split complex objectives into multiple subtasks and assign parallel backend agents for execution. Meanwhile, it maintains dialogue with the user and coordinates the overall workflow. It can spawn ChatGPT Work or Codex subtasks, collect outputs once subtasks finish, append supplementary instructions or bring unresolved judgment points back for human approval.
Completion of a running process does not guarantee the target objective has been fully achieved. OpenAI documentation reminds users to inspect exported artifacts, documents and errors manually. Termination of a runtime process only signals that execution has stopped, and it does not automatically validate correctness or confirm successful delivery.
4. Native Integration within ChatGPT Space
Dots can operate directly on ChatGPT Space pages and comment threads. Users may trigger the agent by typing @ and selecting the target dot, or prefix instructions with @dot. The agent can research page content, draft edits and modify page materials.
Official notes clarify that the Keep Updated function inside ChatGPT Space was unavailable at the Dots launch date. Work requiring periodic refresh still needs separate scheduled tasks created inside chat sessions, with manual validation of timing settings, activation state and runtime results.
Current Access Eligibility for OpenAI Dots
As of September 30, 2026, Dots is releasing in phased rollout. Even users meeting subscription requirements may not immediately see the entry point, as access is deployed incrementally.
OpenAI’s official access matrix covers these tiers:
- Pro 100, Pro 200, Pro 500: Available for users aged 18+, residing within European Economic Area, United Kingdom and Switzerland.
- Business Premium: Global staged rollout.
- Enterprise: Globally available, disabled by default. Workspace administrators must manually enable the feature.
Dots instances are created inside ChatGPT desktop applications or desktop browsers. Once configured, supported mobile app versions allow users to resume conversations with the same dot instance. Mobile web browsers are not supported.
Conversations directly between users and the dot do not consume ChatGPT usage quotas. However, ChatGPT Work and Codex subtasks created or managed by the dot follow standard quota rules for those respective products. Visibility of the Dots interface does not remove rate limits for all backend workloads.
First-time Setup Best Practices for Dot
The priority for initial configuration is not connecting every available tool immediately. Users should define bounded, verifiable long-running responsibilities. The recommended workflow is listed below:
- Launch Dots in desktop browser or ChatGPT desktop client, complete the introductory onboarding flow.
- Connect email, calendars, files and plugins only as required by active tasks. Link a personal local computer only when local file access is necessary.
- Assign a continuous responsibility instead of vague requests such as “help me finish this work”.
- Clearly define source materials, deliverables, refresh frequency, and specify which operations demand mandatory human approval.
- Open the Activity panel inside Dot settings, regularly inspect tasks, generated files, outputs and pending confirmation items.
Sample starting prompt:
> Keep tracking this project continuously. Use the plan and meeting records I provide to maintain action items, deadlines and pending replies. Check progress once each afternoon. Alert me only when deadlines face risk or decisions require my input. You can draft replies, but send nothing before my confirmation.
This instruction clarifies accountability, reference materials, frequency, alert triggers and outbound message boundaries. After validating the first batch of results, users can add integrations for Slack, Teams, Codex or local computers incrementally. This phased approach simplifies troubleshooting when issues emerge.
Will Dot Execute Operations Without Consent or Leak Private Information?
High-risk actions are not enabled for automatic execution by default. Still, Dots cannot replace manual user validation for critical outputs and permission reviews.
Before performing actions that may modify accounts, share content externally or adjust permissions, the system reviews requests against user instructions, permission scopes, custom rules and built-in safety controls. It may pause the workflow, request approval or reject the operation entirely. Password changes or credential modifications are explicitly defined as operations requiring mandatory human confirmation.
Dots offers four customizable rule modes for authorization: allow actions without asking, execute only with explicit user approval, prompt before every execution, or hand all decisions to human users. These custom rules cannot grant permissions for unconnected applications and cannot override the underlying built-in safety guardrails.
Three critical caveats deserve attention:
- Pause only stops the current active task of a dot. It does not shut down all background subtasks.
- Terminating an ongoing task cannot reverse external operations already completed.
- Deleting a dot will not retract messages already sent, nor revert modifications made inside connected applications.
Suitable Workloads for Dots
Dots works best for tasks with changing information, iterative cycles and decision gates requiring human judgment. It is less suited for static, single-input single-output one-off requests.
Common applicable scenarios: continuous sales proposal maintenance, meeting follow-up and milestone deadline tracking, content drafting based on interview transcripts, synchronized document publishing, research data updates, investigating software feedback and preparing code revisions for human audit.
On the other hand, workflows involving irreversible deletions, public releases, fund transfers, sensitive credential changes or production environment edits should never rely solely on broad delegated permissions. For these cases, approval boundaries must be written explicitly into task instructions. Users should inspect background jobs and scheduled runs through Activity and Scheduled panels, and manually sign off on final deliverables.
Conclusion
The core value proposition of OpenAI Dot is not an extra chat interface. It delivers cloud-native persistent agents that carry objectives forward across separate conversations. It links GPT-6 Astra, cloud computing, browser sandbox, backend agent orchestration, ChatGPT Work, Codex and ChatGPT Space into unified long-running workflows, while retaining human oversight and approval gates for high-impact choices.
Dots was officially published on September 29, 2026 during OpenAI DevDay 2026, with phased global rollout. All data in this article is valid up to September 30, 2026. Release regions, supported clients and feature availability may continue to evolve, so final reference should be OpenAI official documentation. For teams managing multi-model agent workflows, an API gateway can simplify unified access and traffic orchestration. 4sapi provides consolidated endpoints to manage different LLM services across development pipelines.
References
- OpenAI DevDay 2026 Release Notes: https://learn.chatgpt.com/docs/whats-new-devday-2026
- OpenAI Docs: Meet dots: https://learn.chatgpt.com/docs/dots
- OpenAI Docs: Get started with your dot: https://learn.chatgpt.com/docs/dots/getting-started
- OpenAI Docs: Tasks and memory: https://learn.chatgpt.com/docs/dots/tasks-and-memory
- OpenAI Docs: Control your dot: https://learn.chatgpt.com/docs/dots/controls
International access: https://4sapi.com
Domestic access: https://4sapi.cn
