Key Notes
- OpenAI Dots bring persistent personal agents powered by GPT-6 Astra into ChatGPT.
- App access, memory retention and approval controls shape how much work users can delegate.
- The rollout starts with one personal dot, while teams of dots remain a future direction.
OpenAI introduced Dots at DevDay 2026 on September 29, bringing persistent personal AI agents into ChatGPT. Powered by GPT-6 Astra, the agents are designed to carry work forward in the background and return with results, questions or requests for approval.
The idea approaches the familiar vision of a personal Jarvis: an assistant that remembers ongoing responsibilities and helps manage them across software. The practical test, however, will be whether Dots can complete useful work reliably while keeping users informed about what they know, what they are doing and which actions require permission.
From Chat Sessions to Ongoing Responsibilities
On its product page, OpenAI describes an agent that can take responsibility for a project across multiple interactions. Users can name their dot, give it a goal and review its progress without keeping a conversation open throughout the job. They can also interrupt, redirect or pause its work.
The company’s examples illustrate how that could change familiar office tasks. A finance assistant follows product and revenue figures, updates an investor presentation and incorporates feedback. A sales assistant compares a customer request with product information before preparing an evaluation plan. An engineering assistant tracks an API migration, prepares changes and tests, and identifies work that remains unfinished.
These are OpenAI’s demonstrations of the intended workflow, rather than evidence that every customer can immediately delegate those jobs successfully. Their common thread is continuity: the agent is expected to retain the assignment as documents, decisions and requirements change.
GPT-6 Astra Supplies the Underlying Capabilities
The underlying model matters because these assignments combine several kinds of work. OpenAI positions GPT-6 Astra for coding, research, browsing and computer use, alongside professional outputs such as documents, spreadsheets and presentations. Those capabilities provide the building blocks for an agent that must investigate a question, use software and deliver an artifact.
OpenAI also describes Astra as better at maintaining a goal when users change direction and at using relevant context during longer tasks. For Dots, that would help an agent revise a presentation after feedback or continue a software project after a requirement changes. Model capability alone, though, does not establish that a particular workflow is dependable.
There are several opportunities for mistakes between understanding an instruction and completing a task: selecting the right document, interpreting a colleague’s message, choosing an action and confirming that it worked. A convincing demonstration therefore leaves a further question for users to test: how much checking does the finished work still require?
Connected Apps Extend the Agent’s Reach
In its announcement, OpenAI says Dots can use a plugin ecosystem covering more than 4,000 apps. The agents can work through ChatGPT and channels including Slack and Microsoft Teams, with connected tools supplying access to relevant services. That ecosystem figure should not be read as automatic access to every user’s accounts.
Each dot has a cloud computer for its work. Connecting a personal computer is a separate choice: OpenAI’s setup guidance says local access is off by default and requires permission. This distinction matters for anyone considering access to work files or applications available on a laptop.
The broader shift toward AI agents is making permissions a central product feature. AIstify’s coverage of Anthropic’s agent controls reflects the same problem: useful automation needs enough access to do its job, together with enforceable boundaries around that access.
Persistent Memory Creates a Separate Set of Choices
According to OpenAI’s privacy FAQ, a dot can receive ChatGPT memories and recent conversation context. Information from conversations with the dot can also contribute back to ChatGPT’s memory. That shared context is intended to reduce repeated explanations of preferences, projects and responsibilities.
The retention rules deserve attention. Disconnecting an app stops new sharing but does not erase information already learned. Turning off ChatGPT memory also does not remove context the dot already holds. At launch, users cannot inspect, edit or delete individual dot memories; clearing that accumulated context requires deleting the dot.
Deleting a dot does not automatically remove other stored material, including Library files, Codex threads, ChatGPT conversations or shared ChatGPT memories. Those items have their own controls. OpenAI says business, enterprise and education data is not used for training by default, while personal users’ treatment depends on their data settings.
For employers, that makes setup a data-management decision as well as a productivity experiment. Teams should understand which records the agent can draw on and where the resulting work will remain.
Background Research Has Limits on Taking Action
OpenAI’s safety overview distinguishes proactive research from actions that change something outside the conversation. Proactive research is restricted to reading: it cannot send messages, edit connected apps or control a browser or desktop. Follow-on actions remain subject to the relevant authorization and approval checks.
Dots run in isolated cloud environments, and users can review activity, redirect work or stop it. Custom rules can define additional boundaries, while a separate action-review system checks proposed operations against the user’s instructions and permissions. Some sensitive operations, such as changing passwords or making financial transfers, are handed back to the user.
The company also describes defenses against prompt injection, in which content in an email, document or website tries to redirect an agent. These protections reduce exposure rather than guarantee an agent will never make a mistake. The more services a dot can reach, the more consequential its interpretation of an instruction can become.
Availability Depends on Plan and Region
The setup guide lists a gradual rollout to eligible Pro and Business Premium users, with Enterprise access in beta requiring an administrator to enable it. Personal Pro access initially excludes the European Economic Area, Switzerland and the United Kingdom. Business Premium availability covers supported ChatGPT regions.
Users must create their first dot through the desktop app or desktop browser. Once set up, it can be used in the mobile app when access is available. Mobile web is not supported for this workflow, and a personal dot does not receive a standalone email address at launch. It also cannot initiate voice calls.
OpenAI says the first dot is included with Pro and Business Premium, with an expanded allowance for deeper work during the first month. Chatting with a dot does not count against ChatGPT limits, but tasks it delegates to Codex or Work use those services’ normal allowances. Users evaluating the cost should distinguish conversation access from the compute consumed by delegated work.
One Personal Dot Now, Broader Teams Later
The initial personal experience centers on one primary dot. OpenAI presents teams of dots as a later direction, while specialist enterprise agents are being tested in pilot programs for functions such as procurement, invoicing and customer support. Those pilots should be distinguished from the personal product becoming available in ChatGPT.
The wider DevDay announcements place Dots alongside changes to ChatGPT, Codex and collaborative work. OpenAI also introduced GPT-6.1 Sol at the event, a separate model announcement from the GPT-6 Astra foundation used by Dots. The collection of releases points toward a service where conversations, software tools and ongoing assignments sit closer together.
For users, the most meaningful measure will be the amount of finished work they can trust after review. A personalized name and persistent context may make an assistant easier to work with. The larger promise of Dots depends on how consistently it turns that familiarity into completed tasks, with clear permissions and a record of what happened along the way.
Disclaimer: AIstify is an independent media brand owned and operated by NuvexMedia LLC, publishing news, research, and insights on artificial intelligence, emerging technologies, automation, and related industries. NuvexMedia LLC invests in and collaborates with companies across the AI, technology, software, and digital innovation sectors. These relationships do not influence AIstify’s editorial coverage, and the publication maintains full editorial independence to provide accurate, timely, and objective information. © 2026 NuvexMedia LLC. All rights reserved. This content is for informational purposes only and should not be considered legal, tax, investment, financial, or other professional advice.
