OpenAI Dots are a new class of always-on AI agents that can keep working between conversations instead of waiting for the next prompt. Announced at OpenAI DevDay on September 29, 2026, each dot is powered by GPT-6 Astra, runs on its own cloud computer, can use a browser and connected apps, and is designed to carry ongoing responsibilities across projects.
The launch marks a meaningful shift in how OpenAI wants people to use ChatGPT. A normal conversation is reactive: you ask, the model answers. A dot is designed to remain active over time, remember the context of an ongoing project, notice relevant changes, continue delegated work, and return when it needs your judgment.
OpenAI says dots can connect to more than 4,000 apps through its plugin ecosystem and can be reached from ChatGPT, Slack and Microsoft Teams, with texting planned. They can also use Codex and ChatGPT Work for delegated tasks, which puts the new product close to the boundary between an AI assistant and a persistent digital worker.
Quick take: OpenAI Dots are rolling out to Pro and Business Premium users, with an Enterprise beta available when administrators enable it. The first dot is included in eligible Pro and Business Premium plans. OpenAI says dots can work 24/7, but consequential actions remain subject to rules, approvals and separate safety checks. OpenAI announced dots on September 29, 2026.
What are OpenAI Dots?
An OpenAI Dots AI agent is a persistent assistant inside ChatGPT that can take responsibility for work that continues after a conversation ends. OpenAI describes dots as “always-on” agents because they can keep making progress in their own cloud workspace while the user is doing something else.
That makes dots different from a standard chatbot session. They are designed to maintain context about goals, standards and ongoing projects, then use that context to decide what work should happen next. OpenAI’s examples include a developer whose dot watches customer feedback and prepares tested fixes, a scientist whose dot reruns analysis as new evidence arrives, and a product team whose dot updates launch materials when requirements change.
The broader idea is familiar from the AI agents category, but OpenAI Dots package several agent capabilities into a persistent consumer and enterprise product: memory, cloud execution, connected tools, delegated work, background research and human approvals.
How the OpenAI Dots AI agent differs from a normal ChatGPT conversation
| Capability | Regular ChatGPT conversation | OpenAI Dots |
|---|---|---|
| Interaction model | Primarily responds when the user starts or continues a conversation | Can continue ongoing responsibilities between conversations |
| Computer | Depends on the ChatGPT feature being used | Each dot has its own cloud computer and browser |
| Long-running work | Usually tied to a specific task or thread | Designed to keep projects moving and revisit them as conditions change |
| Connected apps | Available through ChatGPT plugins and connected apps | Can use connected apps as part of persistent workflows |
| Proactive behavior | Usually user-initiated | Can run restricted proactive research in the background |
| Communication | ChatGPT interface | ChatGPT, Slack, Teams, voice, with texting planned |
| Action controls | Depends on the product and tool | Custom Rules, approvals and Auto-review govern sensitive actions |
OpenAI is not replacing ordinary ChatGPT conversations with dots. Instead, dots are a different operating model for work that has continuity. The user gives the agent a goal and boundaries, then checks progress rather than manually prompting every intermediate step.
OpenAI Dots run on their own cloud computer
One of the defining features is that every dot has its own cloud computer. OpenAI says the dot can browse the web, create files, run tools and use connected accounts inside that protected workspace. The user can inspect the computer and follow what the agent is doing.
The separation matters for both usability and security. The dot does not automatically receive access to the user’s personal computer. Local computer access is optional and starts turned off. If the user explicitly connects a computer through the ChatGPT desktop app, the dot can then work with local files or tools under the permissions granted.
This resembles some of the execution patterns described in Vynula’s OpenAI Agents API guide, where long-running tasks, cloud environments and sandboxed execution separate an agent’s reasoning from the environment in which it acts.
Dots can connect to more than 4,000 apps
OpenAI says its plugin ecosystem gives dots access to more than 4,000 apps. This is strategically important because the usefulness of an always-on agent depends less on its ability to chat and more on whether it can reach the systems where work actually happens.
A dot could, for example, watch project information, retrieve documents, inspect a calendar or use a business application that the user has connected. The exact actions available depend on the app, the permissions granted and the safety rules that apply.
Users manage app permissions through the same ChatGPT connection controls used across ChatGPT, ChatGPT Work and Codex. Disconnecting an app prevents future access through that connection, although information already incorporated into a dot’s context may remain until that context is reset.
What is proactive research in OpenAI Dots?
The most unusual feature may be proactive research. A dot can look for useful information in the background even when the user has not explicitly started a new task.
OpenAI places a significant restriction on this behavior: proactive research uses connected-app tools that are enforced as read-only. The background research process cannot directly send messages, change content inside connected apps, or control the browser or desktop. It can gather information and create private notes for the dot, but any follow-up action must go through the normal action rules.
This distinction matters. An “always-on” agent does not mean the system is free to continuously modify the user’s accounts. The proactive layer is intended to observe and prepare; actions that change external state remain subject to permissions and review.
OpenAI also says it does not directly train its models on the proactive research threads or the dot’s private notes. Information can later enter an eligible conversation or task, however, and its training treatment then depends on the user’s settings and plan.
OpenAI Dots can message users in Slack and Teams
Dots are meant to follow the user across work surfaces rather than remain confined to a single ChatGPT tab. OpenAI says users can message dots through ChatGPT on desktop, web and mobile, as well as through Slack and Microsoft Teams. Voice conversations are also supported, while texting is planned.
The important product idea is continuity. A project could begin in ChatGPT, continue in a team discussion in Slack, and return to the user later with progress without creating a completely separate agent context each time.
OpenAI says dots carry context across these channels. That makes context management especially important; for a deeper technical explanation of persistent agent state, see Vynula’s OpenAI agent memory guide.
Custom Rules define what a dot may do
An always-on agent becomes much more useful when it can take actions, but that also increases the cost of a mistake. OpenAI Dots therefore include a permission system built around default safety rules and user-defined Custom Rules.
Custom Rules can tell a dot to allow, block or seek approval for certain behavior. A user might create a rule that says the dot should never send an email, or that a certain class of action must always be confirmed first.
These rules cannot remove OpenAI’s mandatory safety requirements. OpenAI says some categories of action always require explicit confirmation or must be handed back to the user.
| Action type | How OpenAI says dots handle it |
|---|---|
| Routine permitted work | May proceed when covered by the user’s instructions and rules |
| Sending or sharing information | Authorization depends on the information and recipient |
| Purchases with a saved card | User approval required |
| Permanent data deletion | Confirmation required each time |
| Installing software from an unrecognized source | Confirmation required |
| New security-sensitive access | Confirmation required |
| Password changes | Dot must hand the sensitive step back to the user |
| Financial-account transfers | Dot must hand the sensitive step back to the user |
Auto-review checks actions before they happen
OpenAI also describes a separate safety mechanism called Auto-review. Before a dot carries out actions such as sending an email or changing a file, Auto-review evaluates the planned action against the user’s instructions, Custom Rules and OpenAI’s mandatory safety requirements.
If the action passes, the dot can proceed. If Auto-review blocks it, the action does not run. The dot receives the reason and can ask the user for more context or approval where appropriate, attempt a permitted alternative, or stop.
OpenAI says the controls enforcing Auto-review sit outside the environments that dots can modify. The goal is to prevent the agent from disabling the mechanism that is supposed to supervise it.
That is an important design choice for an autonomous system. It follows the same general principle discussed in Vynula’s broader coverage of AI agent guardrails: a rule is more meaningful when it is enforced outside the component whose behavior it constrains.
How passwords and account sign-ins work
OpenAI says dots can sign into supported websites without exposing passwords to the model. During secure sign-in, the model is paused while the user completes the login, and credentials are sent directly to the browser environment rather than inserted into the conversation context.
Supported saved-password flows use a dedicated encrypted credential service. The dot can use the credential for an authorized sign-in, but the password itself is not passed into the model’s context.
This does not mean every secret is automatically protected. OpenAI explicitly notes that a password or API key placed in a normal readable message or document may still be visible to the model. The protection applies to the dedicated secure sign-in and saved-password mechanisms.
What are specialist dots?
The consumer-facing product starts with a personal primary dot, but OpenAI is also previewing specialist dots for organizations.
A specialist dot would have its own enterprise identity, credentials and tightly scoped access to systems needed for a defined job. OpenAI says it has already tested this model internally across areas including procurement, invoice processing, email marketing, customer support and commercial contracting.
The company is beginning with focused enterprise pilots rather than making specialist dots broadly self-service on day one. OpenAI says its engineering teams will work with participating organizations to define responsibilities, tools and approval workflows.
This is where the product begins to overlap with the concept of AI employees or digital workers: the agent is not simply completing an isolated task, but is assigned a recurring organizational responsibility.
Microsoft Agent 365 will help govern specialist dots
OpenAI says it is working with Microsoft to integrate specialist dots with Microsoft Agent 365. The stated goal is to allow enterprise IT teams to manage dots through Microsoft’s existing governance and security controls.
That integration is significant because enterprise adoption will depend on identity, credentials, access policies, auditing and administrative controls at least as much as model capability. A company may be willing to let an autonomous agent work continuously only if the agent can be treated as a governed identity rather than an invisible background process.
Who can use OpenAI Dots today?
The initial rollout differs by plan and region.
| Plan | Initial availability |
|---|---|
| Pro | Rolling out in eligible markets; OpenAI’s Help Center currently excludes the European Economic Area, Switzerland and the UK for Pro access |
| Business Premium | Available across supported ChatGPT regions, with gradual rollout |
| Enterprise | Beta available when the workspace administrator enables it; off by default |
| Edu and Healthcare | Included in the Enterprise-style beta when enabled by an administrator |
OpenAI says the first dot is included in Pro or Business Premium at no additional cost. Those plans also include an allowance for deeper work, with temporarily extended limits after launch. The company says users will eventually be able to add additional dots and increase the speed or monthly work capacity of individual dots.
Normal conversations with a dot do not count against ChatGPT usage limits, according to OpenAI. When a dot starts work in Codex or ChatGPT Work, those delegated tasks use the corresponding product limits as usual.
What can an OpenAI Dot actually do?
The launch examples show the kind of ongoing responsibility OpenAI is targeting:
- Software development: monitor feedback, identify recurring requests, implement smaller fixes, test them and prepare pull requests for review.
- Product launches: update plans, messaging and assets as scope changes.
- Research: rerun analysis when new evidence appears and update figures or explanations.
- Sales: compare customer requirements with product documentation, prepare proofs of concept and keep proposals current.
- Content production: watch for new source material, prepare clips, notes and social drafts, then incorporate editorial feedback.
These examples are OpenAI’s descriptions of intended use, not independent evidence that every workflow will operate reliably without supervision.
The DevDay demo also showed why supervision still matters
OpenAI presented dots live at DevDay, but the demonstration was not flawless. Reuters reported that some live voice-update demonstrations repeatedly failed to return the expected updates, prompting an on-stage acknowledgment of technical problems.
That does not establish that the underlying product is unreliable, but it is a useful counterweight to the marketing promise of an agent that works continuously. Persistent agents introduce more opportunities for small errors to accumulate: misunderstood goals, stale context, unexpected app behavior, permission problems and failures in long-running tasks.
OpenAI itself repeatedly warns that dots can make mistakes and recommends reviewing consequential work. The product therefore should not be interpreted as a promise that a user can delegate any high-impact workflow and stop monitoring it.
Safety is a core part of the OpenAI Dots launch
Safety is unusually prominent in OpenAI’s product material for dots. That reflects the larger risk surface created when an AI system can browse, use accounts, write files, send messages and keep working without the user continuously watching.
OpenAI says GPT-6 Astra supplies the model-level behavior, while additional protections operate around the model: isolated workspaces, secure sign-ins, tool restrictions, monitoring, Custom Rules, approvals and Auto-review.
Prompt injection remains a concern. A malicious webpage, email or document can contain instructions intended to redirect an agent or induce it to reveal information. OpenAI says its defenses combine model safeguards with restrictions and checks before actions. The company can also pause work if monitoring detects potentially harmful behavior.
The existence of these safeguards should not be treated as proof that prompt injection or autonomous-agent failures are solved. OpenAI’s own safety documentation describes remaining limitations and explicitly says dots can still make mistakes.
OpenAI Dots vs ChatGPT Work
Dots and ChatGPT Work overlap, but they are not the same product concept.
ChatGPT Work is a way to hand off substantial tasks that may require research, files, apps, browsing and multi-step execution. A dot adds persistent identity and continuity: it is designed to know the user’s ongoing priorities, remain available, continue working between conversations and proactively notice opportunities to help.
A dot can also delegate work into ChatGPT Work or Codex. In that sense, the dot acts more like a persistent coordinator while Work and Codex can be execution environments for specific jobs.
For developers, this is also different from directly building with the OpenAI Agents API. The API is infrastructure for creating agentic software; dots are an OpenAI product that packages persistent-agent behavior for end users and organizations.
Vynula analysis: the assistant is becoming a standing worker
The most important change in OpenAI Dots is not the cloud computer, Slack integration or even the 4,000-app ecosystem by itself. It is the shift in responsibility over time.
Traditional assistants wait for work. The user notices a problem, opens the app, explains the task and requests an answer. OpenAI Dots reverse part of that flow: the user defines an ongoing responsibility and a set of boundaries, then the agent is expected to keep the problem in mind.
That creates a new product category somewhere between chatbot, automation platform and digital employee. The agent can notice new information, decide that it matters to an existing goal, prepare work and contact the user when judgment is required.
This could reduce coordination overhead for people who manage many small, changing tasks. It could also create a new management problem: users may have to supervise not just individual AI outputs, but a portfolio of agents with persistent context, app access and ongoing responsibilities.
OpenAI is already signaling that future users may have multiple dots and that organizations may deploy specialist dots with independent identities. If that model grows, the important enterprise questions will include who owns an agent, who approves its permissions, how its performance is evaluated, what happens when it changes roles, and how accumulated context is governed.
In Vynula’s view, that is the real significance of the launch: OpenAI is trying to turn AI from a tool you invoke into a worker you manage. Whether users actually trust dots with that level of responsibility will depend less on impressive demos and more on reliability, transparent controls and evidence that autonomous work remains inside the boundaries people intended.
What to watch next
- Real-world reliability: how well dots handle projects that evolve over days or weeks.
- Approval fatigue: whether safety prompts remain useful without becoming so frequent that users automatically accept them.
- Regional expansion: when Pro access reaches currently excluded European markets.
- Specialist dots: whether enterprise pilots turn into a broad product for persistent organizational agents.
- Microsoft governance: how deeply Agent 365 can manage dot identity, permissions and auditability.
- Agent economics: how OpenAI prices additional dots, faster execution and larger monthly workloads.
- Security incidents: whether prompt injection, credential misuse or incorrect autonomous actions expose limits in the safeguards.
Frequently asked questions
What are OpenAI Dots?
OpenAI Dots are always-on AI agents in ChatGPT that can keep working across ongoing projects between conversations. They use GPT-6 Astra, have their own cloud computer and can work with connected apps under user-defined permissions.
What model powers OpenAI Dots?
OpenAI says dots are powered by GPT-6 Astra.
Can OpenAI Dots use my computer?
Yes, but local computer access is optional and starts disabled. Each dot has its own cloud computer by default. Users can explicitly connect a personal computer through the ChatGPT desktop app.
Can Dots act without asking permission every time?
Yes for some actions that fit the user’s instructions and rules. Other actions require approval, and certain sensitive steps—such as changing a password or transferring money between financial accounts—must be handed back to the user.
Can OpenAI Dots work in Slack and Microsoft Teams?
Yes. OpenAI says dots can communicate through ChatGPT, Slack and Teams. Text messaging is planned.
How many apps can OpenAI Dots connect to?
OpenAI says dots can readily connect to more than 4,000 apps through its plugin ecosystem.
Are OpenAI Dots available in Europe?
Availability depends on the plan. OpenAI’s current Help Center says Pro access excludes the European Economic Area, Switzerland and the UK at launch, while Business Premium is available across supported ChatGPT regions. Enterprise, Edu and Healthcare workspaces can access the beta when administrators enable it.
Primary and independent sources
- OpenAI — Introducing dots, September 29, 2026
- OpenAI — How we build safety, security, and privacy into dots
- OpenAI Help Center — Getting started with your dot
- OpenAI — DevDay 2026 recap
- Reuters — OpenAI takes on Meta with dots agent in autonomous AI push
Reporting note: Product capabilities, availability and safety mechanisms are attributed to OpenAI unless otherwise stated. Reuters is used for independent event context, including the live demo issues. Vynula Analysis represents editorial interpretation. Featured and inline images are illustrative Unsplash photography, not official OpenAI product imagery.
