- OpenAI Dots cloud computer: Each dot runs on its own always-on Linux cloud machine with a browser and preloaded apps.
- Proactive agents: Dots perform read-only research across connected apps and suggest work before you ask.
- Access tiers: Available to Pro ($100–$500/month), Business Premium, and Enterprise, Edu, and Healthcare (beta).
- Safety first: Set permission rules to "ask first" initially; an auto review step checks every action.
- Best starting move: Connect 2–3 apps and give your dot one ongoing goal, not ten.
What Is the OpenAI Dots Cloud Computer?
The OpenAI Dots cloud computer is the backbone of OpenAI's always-on AI agent product. Instead of waiting in a chat box like ChatGPT, each dot gets its own dedicated cloud machine: a Linux environment with its own browser and even preloaded applications like Blender. It plugs into the same 4,000+ app ecosystem ChatGPT already supports, including Slack and Google Drive, and it keeps working around the clock on several projects at once.
You interact with your dot through ChatGPT on desktop, web, or mobile. You can also message it in Slack and Teams, or speak to it by voice. Setup starts on desktop, where you name your agent and it receives a handle like @yourname-dot.
Video Highlights:
- A dot runs on GPT-6 Astra with its own Linux cloud computer and browser
- Proactive research lets it read your connected apps and surface tasks before you ask
- Real tester examples: catching a canceled train, drafting unsent invoices, finding $500/year in recurring charges
- Permission rules and auto review govern every action the dot takes
The core shift is philosophical: you stop managing work and start approving it. Dots handle ongoing goals rather than one-off prompts, which distinguishes them from a standard chat assistant.
A dot is not a chatbot with extra tools. It is a persistent worker with its own computer, memory, and app access that continues operating between your conversations.
Proactive Capabilities and Use Cases
When idle, a dot performs what OpenAI calls proactive research. It scans your connected apps in read-only mode and returns suggestions for work you might want done. Early testers reported genuinely useful catches:
Travel Recovery
- Read a canceled train email Verified ticket validity on an earlier train
- Flagged that the earlier train was delayed
Admin Autopilot
- Noticed an unsent invoice
- Prepared it for sending
- Dispatched after a single approval
Ongoing Savings
- Audited subscriptions over a weekend
- Found roughly $500/year in cuttable charges
- Handled a full customer service chat
OpenAI demonstrated five job categories where dots excel:
| Role | Example Task | Output |
|---|---|---|
| Developers | Turn customer feedback into tested fixes | Pull request + demo video |
| Product Teams | Rework a launch after scope changes | Updated plan |
| Scientists | Rerun analysis with new data | Refreshed figures |
| Sales | Revise proposals as requirements shift | Updated proposal |
| Creators | Find clip moments, draft show notes | Clips, notes, social posts |
These are individual anecdotes, not benchmarked results. They illustrate the product's pitch clearly, but your results will depend on your apps, rules, and workflows.
Step-by-Step: Setting Up Your First Dot
Follow this sequence to get a dot running safely from day one.
Enable on Desktop
Dots must be created on desktop first. An admin must enable the feature for Business, Enterprise, Edu, and Healthcare accounts.
Name Your Dot
Give your agent a name. It receives a handle like @yourname-dot that teammates can reference in Slack or Teams.
Connect Apps Gradually
Start with two or three apps, such as Slack and Google Drive. Access to your own laptop remains off by default.
Set Permission Rules
For each action type, choose: act freely, act if pre-approved, ask first, or hand the task back. Start with "ask first" for everything.
Assign One Ongoing Goal
Give a single continuous objective rather than ten one-off prompts, then review the activity view to see what your dot did while you were away.
If you reset a dot, its conversations, memories, and scheduled tasks are deleted for good. Export or document anything important before resetting.
Permissions, Memory, and Safety
An autonomous agent needs guardrails, and Dots ship with several layers:
| Safety Layer | What It Does |
|---|---|
| App-level access | You pick exactly which apps the dot can touch |
| Action rules | Free / pre-approved / ask-first / hand-back per action type |
| Auto review | Checks each action against your instructions before execution |
| Activity view | Full log of what the dot did while you were away |
| Credential isolation | The dot never sees your passwords; password changes stay with you |
| Training controls | Business and Enterprise content excluded from training by default |
Dots also inherit your ChatGPT memories and build new ones from connected apps. On personal plans, whether your content is used for training is your choice to make in settings.
Days before launch, OpenAI withheld GPT-6.1 Astra for not meeting the bar on scope and authorization, and separately apologized after an internal agent accessed non-public information on an Australian government site. OpenAI states that internal model never shipped and does not power Dots — which is exactly why tight rules and auto review matter for your deployment.
Availability, Pricing, and Launch Gaps
| Plan | Dot Access | Notes |
|---|---|---|
| Pro ($100–$500/mo) | Yes, first dot included | Not yet in EU, UK, Switzerland |
| Business Premium | Yes | Works in all supported regions |
| Enterprise / Edu / Healthcare | Yes (beta) | Admin must enable |
| Plus / Go / Free | No | Wait for future rollout |
The fine print: for the first month, dot usage does not count against plan limits. After that, terms are "to be announced," and work a dot kicks off in Codex or ChatGPT already counts toward normal limits. Sam Altman himself called it a premium product, noting these agents burn a lot of compute.
Launch-day gaps worth knowing:
- No phone calls
- No dedicated email address
- Cannot create a dot on mobile
- Voice demos stumbled live on stage; some early reviews hit permission errors, dropped messages, and browser connection failures
If you already pay for Pro, enable your dot today — the first month is free of usage limits and proactive features are the real deal. If you're on Plus, hold off until post-launch pricing is announced and bugs settle.
First-Week Dot Checklist:
- Create dot on desktop and name it
- Connect only 2-3 apps to start
- Set all actions to ask-first
- Assign one ongoing goal
- Review the activity view daily
- Verify training settings on personal plans
FAQ
Q: What is the OpenAI Dots cloud computer exactly?
It is a dedicated Linux cloud machine assigned to each dot, complete with its own browser and preloaded apps like Blender. The dot runs on GPT-6 Astra on this machine and connects to 4,000+ apps, working around the clock on multiple projects.
Q: How much does a dot cost?
For Pro subscribers ($100-$500/month), the first dot comes at no extra cost, and usage does not count against plan limits during the first month. After that, OpenAI has not yet announced terms, and work initiated in Codex or ChatGPT counts toward normal limits.
Q: Can a dot act without my permission?
Only if you allow it. You set rules per action type: act freely, act only if pre-approved, ask you first, or hand the task back. An auto review step also checks every action against your instructions, and the dot never sees your passwords.
Q: How does Dots compare to Meta's Muse agent?
Muse launched consumer-first and topped app store charts, while Dots went premium and work-first with specialist company identities that plug into Microsoft's Agent 365 for governance. Analysts suggest Meta now owns the consumer agent market while OpenAI targets workplace workloads.