OpenAI launched Dots on September 29, 2026, with a large promise: an assistant that keeps working while you are away. Each dot has its own cloud computer, browser, memory, and access to connected apps. You can also give it access to your laptop. The useful question is how much work it can complete safely, and what happens to its accumulated context when you change your mind.
This breakdown covers the launch documentation, pricing and access, privacy controls, early reactions, and alternatives. The keynote showed what OpenAI wants Dots to become. Its launch post and getting-started guide establish what buyers can actually evaluate today.
What Is OpenAI Dots?
A dot is a named, persistent AI assistant inside ChatGPT, powered by GPT-6 Astra. OpenAI gives it a hosted computer and browser so it can work across tasks and projects, including when your own computer is closed. It can connect to more than 4,000 apps through OpenAI's plugin ecosystem and carry context between ChatGPT, Slack, and Microsoft Teams. You can inspect the dot's computer as it works. OpenAI is starting users with one primary dot; teams of specialist dots remain an enterprise pilot or future plan, not a standard multi-dot subscription feature. OpenAI announcement.
How OpenAI Dots Works
- A cloud computer is the home base. A dot has a browser and computing environment hosted by OpenAI. That is where its ongoing work runs. OpenAI announcement.
- Laptop access is optional. With the ChatGPT desktop app, you can separately grant access to local files, skills, and browser tasks. It starts off. Dots are cloud-hosted, but saying they cannot work on a local machine is wrong. Getting-started guide.
- Proactive research is read-only. While looking for useful work in connected apps, the dot cannot send messages, change app content, or control a browser or computer. User-directed and scheduled tasks can act under separate rules and approvals. OpenAI announcement · Safety explainer.
- Memory follows the dot. It can use ChatGPT memories and build its own from conversations and connected apps. The same continuity that makes it useful also raises a retention question when an app is disconnected. Getting-started guide.
The boundary is clearer in two launch examples. In OpenAI's account, an early tester's dot noticed a missed invoice, prepared it, and sent it after approval. In a dinner demo reported by WIRED, a dot saw an evening calendar conflict and offered two GrubHub options with prices; the user then chose the food and when to order. It did not independently place an order in that example. Both illustrate the intended workflow, not a measured real-world success rate.
OpenAI Dots Pricing and Access
How much does a dot cost?
OpenAI says the first dot is included at no extra charge with eligible ChatGPT Pro and Business Premium plans. Enterprise, Edu, and Healthcare workspaces can try an admin-enabled beta. Access is rolling out by market, and Pro availability excludes the EEA, UK, and Switzerland at launch. Conversations with a dot do not consume ordinary ChatGPT usage, but tasks it starts in Codex or ChatGPT Work count against those products' limits. Its plan includes a deeper-work allowance with extended limits during the first month. OpenAI has not published the ongoing Dots allowance or pricing for additional dots. OpenAI announcement · Getting-started guide.
There is a related subscription change, but it should not be mistaken for a published Dots price. On September 29, OpenAI's Tibo Sottiaux said the $200 Pro subscription would reopen to new subscribers on September 30 with a new usage calculation. In his comparison, the subscription would represent about half the API-dollar equivalent of the old Pro plan. He said the five-hour usage limit would not return and hinted at benefits that would not draw on usage. This is a change to the wider Pro economics, not a disclosed post-introductory allowance for Dots. Buyers should wait for the actual Dots limits before calculating the cost of unattended work.
Setup starts in a desktop browser or desktop app; subsequent conversations are available on supported mobile clients. You can call your dot, but it cannot initiate a call to you at launch. Texting is a limited US Pro beta, not a general communication channel. OpenAI announcement · Getting-started guide.
OpenAI Dots Security and Privacy Controls
OpenAI describes built-in action rules, user-defined Custom Rules, automatic review for actions affecting accounts or sharing information, and monitoring that can pause suspicious work. Some sensitive tasks, including changing a password, remain with the user. Supported saved-password flows keep credentials out of the model's context. These are meaningful design controls; their failure rates in real-world use have not yet been independently established. OpenAI safety explainer.
The retention rule deserves a pause before you connect a sensitive account. Disconnecting an app does not erase information the dot already learned from it. Resetting the dot deletes its conversations, memories, and scheduled tasks together. There is a difference between revoking future access and removing past context, and OpenAI's getting-started guide makes it explicit. For personal plans, training use depends on your settings; OpenAI says Business, Enterprise, and Edu workspace content is not used for model improvement by default. OpenAI announcement.
What OpenAI Dots Does Well
- Work can continue away from your laptop. Its hosted computer gives tasks a persistent place to run. OpenAI announcement.
- Local access is a choice rather than a blanket permission. Users can grant the desktop connection when a task needs it. Getting-started guide.
- The app ecosystem is broad. OpenAI says its plugins connect to more than 4,000 apps. OpenAI announcement.
- Approval boundaries are documented. Custom Rules, auto-review, and Activity View give users ways to watch and constrain action. OpenAI safety explainer.
Where OpenAI Dots Falls Short
- The ongoing economics are unknown. Introductory limits do not establish next month's cost per completed task. OpenAI announcement.
- The primary runtime is closed and hosted by OpenAI. Optional laptop access does not provide a self-hosted dot or model-provider choice. OpenAI announcement.
- Data removal is coarse. Disconnecting a source does not remove what the dot learned, while resetting it also erases schedules and other memories. Getting-started guide.
- The channel story is still developing. The dot cannot initiate calls at launch, and texting is limited. Getting-started guide.
- Long-run reliability is unproven. Launch demos and short early-access accounts cannot establish weeks of correct, permissioned work.
The Personal-Agent Market Is Moving Fast
Meta's Muse shows demand for this category: Semafor reported more than 3 million downloads by the day Dots launched. Downloads show interest, not sustained task success. Adoption has also exposed practical limits. Amazon blocked Muse from shopping on its site, illustrating that an agent's ability to reach a service depends on that service allowing it. In a separate Facebook Marketplace incident investigated by The Guardian, Muse sent a buyer a user's home address after the user enabled automatic replies; he said he had not approved sharing the address. He later said the scope of the "Allow Always" permission was not clear to him. These are Muse-specific cases, not evidence of a Dots failure. They make site access and comprehensible approval boundaries concrete questions for anyone evaluating an always-on agent.
Who Should Consider OpenAI Dots?
- ChatGPT-heavy teams: You already use OpenAI products and want one agent to follow ongoing projects.
- People who want managed setup: You prefer OpenAI to operate the computer and integrations.
- Teams testing governed delegation: You can pilot narrowly and review consequential actions before expanding access.
Who Should Look at Alternatives?
- People who want to run the primary assistant themselves: A laptop connection is different from choosing the runtime.
- Teams that need model choice: Dots runs on OpenAI's model stack.
- Buyers who require predictable task economics today: The ongoing Dots allowance is undisclosed.
- People with strict deletion requirements: Test whether the reset behavior fits your retention policy.
- Anyone needing proven long-term performance: Wait for independent multi-week use, not a keynote highlight reel.
Best OpenAI Dots Alternatives (2026)
1. Vellum
Vellum is an open-source personal AI assistant available through a native desktop app, Vellum Cloud, or self-hosting. It keeps context across desktop, iOS, Android, web, voice, email, Telegram, and Slack, and supports multiple model providers. Its documentation and public codebase let buyers inspect the deployment and permission model.
Standout strengths: Local or managed primary runtime; open-source ownership; model selection; memory across surfaces; an extensible skill and plugin system.
Trade-offs: Self-hosting takes maintenance; a personalized assistant needs time and corrections to learn your preferences.
Compared to Dots: Both can act across apps, remember context, and work with a laptop. Vellum lets you choose where the assistant primarily runs and which model handles a task. Dots offers the simplest path for a buyer committed to OpenAI's hosted stack.
2. Claude Cowork
Claude Cowork focuses on delegated document, file, and application work in Anthropic's environment. Its strongest fit is an existing Claude workflow; it does not diversify the model provider beyond Anthropic. Anthropic product page.
Standout strength: Focused knowledge-work tasks. Trade-off: Less directly aimed at the same cross-channel personal-assistant use case.
3. OpenClaw
OpenClaw is an open-source, self-operated agent for technical users. Hosting and configuration choice are the draw; maintaining permissions, updates, and secrets becomes your job. Project repository.
Standout strength: Hands-on ownership. Trade-off: More operational responsibility than a managed dot.
4. Perplexity Computer
Perplexity Computer is worth evaluating for research and delegated analysis. Compare its current task and scheduling scope directly before treating it as a complete replacement for a persistent personal assistant. Perplexity product page.
Standout strength: Research-oriented workflows. Trade-off: Different center of gravity from ongoing personal context.
5. Meta Muse
Meta Muse is a managed personal agent tied to Meta's services and messaging ecosystem. The relevant decision is which company should operate the environment and hold the context. Meta announcement.
Standout strength: Managed consumer-agent experience. Trade-off: Another closed ecosystem rather than a portable runtime.
These are editorial fits, not laboratory rankings. We have not tested five agents on a common workload, so numerical scores would pretend to measure something we have not measured.
OpenAI Dots Alternatives Comparison
| Assistant | Primary Environment | Best Fit | Main Trade-off |
|---|---|---|---|
| Vellum | Local app, Vellum Cloud, or self-hosted | Ownership and model choice | Self-hosting requires maintenance |
| OpenAI Dots | OpenAI-hosted computer; optional laptop connection | Managed ongoing work in ChatGPT | Ongoing Dots allowance undisclosed |
| Claude Cowork | Anthropic-managed work with desktop access | Files and knowledge work | Anthropic ecosystem |
| OpenClaw | Self-operated | Technical owners | Security and operations are yours |
| Perplexity Computer | Managed service | Research-led tasks | Different personal-context emphasis |
| Meta Muse | Meta-managed | Consumer-agent tasks | Closed ecosystem |
Vellum vs. OpenAI Dots
Where does the assistant run?
A dot's primary computer is OpenAI-hosted, even when you authorize tasks on a laptop. Vellum gives you a native local route, managed cloud route, or self-hosted route. The distinction is the home base, not whether either product can touch a desktop.
Can you change the model?
Dots is built on GPT-6 Astra. Vellum's public repository documents support for several providers and local models. That flexibility matters when pricing, capability, or data-handling requirements change.
Who controls accumulated context?
OpenAI's reset behavior creates an all-or-nothing deletion decision for a dot. Vellum's docs describe export and deletion of workspace data, alongside local or managed hosting. Either way, connect only the accounts your actual tasks need.
What Early Users Are Saying
The DevDay keynote discussion and launch thread are reactions, not longitudinal reviews. Questions center on permission mistakes, usage, and regional access. Dan Shipper's public early-access excerpt gives one more concrete observation: his dot flagged a calendar conflict based on a Slack discussion he had missed. The warning was grounded, not a hallucination. Most of his review is paywalled, so it cannot support a broader verdict here. No first-day post can answer whether a dot stays useful after weeks of real work.
FAQs
Is OpenAI Dots available to everyone?
No. OpenAI is rolling it out to eligible Pro and Business Premium users by market, with an admin-enabled Enterprise beta. Pro access excludes the EEA, UK, and Switzerland at launch. Getting-started guide.
Can a dot use my laptop?
Yes, if you enable the optional desktop connection. The dot's primary computer remains hosted by OpenAI. Getting-started guide.
Does proactive work let it send messages without asking?
Proactive research is read-only. Other requested or scheduled tasks may act under rules and approval gates. OpenAI safety explainer.
Can a dot call me?
You can call your dot, but it cannot initiate a call to you at launch. Getting-started guide.
Can I text my dot?
Texting is limited to a US Pro beta where available, with broader access planned. Do not assume it is enabled on every account. OpenAI announcement.
Do dot conversations consume ChatGPT usage?
OpenAI says dot conversations do not count against ordinary ChatGPT usage; tasks started in Codex or ChatGPT Work count under those products' limits. OpenAI announcement.
How much will deeper Dots work cost after the first month?
OpenAI has not published the ongoing allowance or a per-task price. The wider Pro subscription change does not answer that Dots-specific question. OpenAI announcement · Tibo Sottiaux.
What happens when I disconnect an app?
The dot loses future access, but OpenAI says it may retain information already obtained from the app. Resetting the dot deletes conversations, memories, and schedules together. Getting-started guide.
Is OpenAI Dots open source or self-hosted?
No. Its primary environment is managed by OpenAI; optional access to your laptop does not make the dot self-hosted. OpenAI announcement.
Is Vellum only for people who self-host?
No. Vellum offers a managed cloud route and a native desktop app as well as self-hosting. Vellum docs.
Which assistant offers more model choice?
Vellum supports multiple cloud providers and local models. Dots is built on OpenAI's GPT-6 Astra. Vellum repository · OpenAI announcement.
Have independent reviewers established Dots' reliability?
Not over a meaningful period yet. Early-access anecdotes can illustrate a workflow; they cannot establish a failure rate or long-term trustworthiness. Dan Shipper's excerpt.
What should I test before connecting sensitive accounts?
Start with one limited task, confirm approval stops where expected, check what activity you can inspect, and read the reset rules before granting broad access. OpenAI safety explainer · Getting-started guide.
Extra Resources
- What Can a Personal AI Assistant Do?
- Official Claude Cowork Breakdown (2026)
- Claude Opus 5.5 vs GPT-6 Astra
The product architecture is documented. Its long-run reliability and Dots-specific post-introductory economics are not. Those are the two things worth checking before handing an always-on agent your busiest accounts.
