- OpenAI Dots proactive agent is an always-on AI coworker running on GPT-6 Astra with its own cloud computer.
- Proactive research: A dot scans connected apps read-only and surfaces tasks before you ask.
- Access: Pro, Business Premium, Enterprise, Edu, and Healthcare (beta) users, once an admin enables it.
- Safety first: Set action rules and auto review tightly from day one.
- First dot included: Pro users get their first dot at no extra cost, but usage terms after month one are TBD.
What the OpenAI Dots Proactive Agent Actually Is
The OpenAI Dots proactive agent is a fundamentally different product from ChatGPT. Instead of waiting in a chat box for a prompt, a dot runs continuously on GPT-6 Astra with its own dedicated cloud computer — a Linux machine with a browser and even preloaded apps like Blender. It plugs into the same 4,000+ apps ChatGPT already connects to, including Slack and Google Drive, and works on several projects simultaneously around the clock.
Video Highlights:
- A dot is an always-on agent running on GPT-6 Astra, not a chat session
- Each dot gets its own cloud Linux computer with a browser and preloaded apps
- It connects to 4,000+ apps and keeps working across multiple projects
- You interact via ChatGPT on desktop, web, phone, Slack, Teams, or voice
The defining feature is proactive research. When you haven't assigned anything, a dot looks through your connected apps (read-only) and returns suggestions for work you might want done. The core pitch: you stop managing the work and start approving it.
| Feature | Regular ChatGPT | Dots Proactive Agent |
|---|---|---|
| Runtime | On-demand chat | Always-on, 24/7 |
| Compute | Shared session | Own cloud Linux computer |
| App access | Connects on request | 4,000+ apps continuously monitored |
| Initiation | You prompt it | Proactive research between tasks |
| Interaction | Chat interface | Chat, Slack, Teams, voice |
Each dot gets a name and a handle like @yourname-dot. Setup must be done on desktop first — you cannot create a dot from your phone, though you can message it from any device afterward.
Real Proactive Use Cases from Early Testers
Early hands-on reports show what proactive behavior looks like in practice. These are individual experiences rather than formal measurements, but they illustrate the pattern clearly.
Travel Recovery
- Train canceled overnight
- Dot read cancellation email unprompted
- Verified ticket on earlier train
- Caught that earlier train was late
- Tester made his connection
Invoice Follow-Up
- Forgotten invoice detected
- Dot prepared it in advance
- Sent once tester approved
- Zero manual drafting
Cost Audit
- ~$500/year in savings found
- Scanned recurring charges over a weekend
- Handled a customer service chat end-to-end
OpenAI demonstrated five job categories where dots shine, all sharing one pattern — they're ongoing goals, not one-off prompts:
| Role | Ongoing Job Example |
|---|---|
| Developers | Turn customer feedback into tested fixes with a pull request and demo video |
| Product teams | Rework a launch when scope changes |
| Scientists | Rerun analyses when new data arrives, update figures |
| Sales | Revise proposals as requirements shift |
| Creators | Find clip moments in transcripts, draft show notes and social posts |
One reviewer reported his dot flagged that a flight change clashed with a Slack meeting request he hadn't read yet — exactly the kind of cross-app awareness proactive agents are built for.
Not every tester is sold. One developer noted dots are very proactive with connected apps from day one but weak at writing, describing output quality as "very GPT-4." A longer review hit permission problems, dropped messages, and a flaky browser connection.
Permission Rules and Safety Settings
An autonomous agent needs brakes. Dots give you layered control over what your agent can touch and how it acts.
| Action Setting | Behavior |
|---|---|
| Act freely | Dot acts without asking for this action type |
| Pre-approved only | Dot acts only if you preapproved it |
| Ask first | Dot requests permission every time |
| Hand back | Dot returns the task to you |
Additional safeguards:
- App-level access: You pick exactly which apps a dot can touch.
- Auto review: Every action is checked against your instructions and rules before execution.
- Activity view: See everything your dot did while you were away.
- Password isolation: Dots never see your passwords; sensitive actions like password changes always stay with you.
- Laptop access off by default: Access to your local machine is opt-in.
The week of launch, OpenAI disclosed an agent incident involving non-public data on an Australian government website (attributed to an internal model, not Dots), and held back GPT-6.1 Astra for not meeting the bar on scope and authorization. Set your dot's rules tight from the start — do not default to "act freely."
Memory and privacy deserve attention too. A dot pulls in your ChatGPT memories and builds its own from connected apps. On Business and Enterprise plans, your content isn't used for training by default; on personal plans, that's your choice. Resetting a dot is permanent — conversations, memories, and scheduled tasks are gone for good.
Step-by-Step: Setting Up Your First Dot
Verify Plan Access
Confirm you're on Pro, Business Premium, Enterprise, Edu, or Healthcare beta. Plus, Go, and free users don't have access yet, and Pro users in the EU, UK, and Switzerland are still waiting.
Create on Desktop
Open ChatGPT on desktop to create your dot. Give it a name, and it receives a handle like @yourname-dot.
Connect Two or Three Apps
Start small — connect Slack, Google Drive, or another core tool. Don't grant access to everything on day one.
Set Action Rules to 'Ask First'
Configure each action category conservatively. Start with ask-before-acting and loosen permissions only after the dot earns trust.
Assign One Ongoing Goal
Give it a single ongoing objective — invoice follow-ups, weekly recaps, proposal tracking — not ten tasks. Review the activity view daily at first.
First-Week Milestones:
- Create dot on desktop and name it
- Connect 2-3 core apps only
- Set all action rules to 'ask first'
- Review activity view each morning
- Assign one ongoing goal, not ten
For the first month, your dot's usage doesn't count against your plan limits. Use this window to experiment — but remember that post-month-one terms have not been announced, and work a dot kicks off in Codex or ChatGPT already counts toward normal limits.
Pricing, Availability, and Fine Print
Sam Altman himself called dots a premium product, noting they burn a lot of compute. Here's the access picture as of late September 2026:
| Plan | Access | Notes |
|---|---|---|
| Pro ($100–$500/month) | Yes | First dot included at no extra cost |
| Business Premium | Yes | Works in all supported regions |
| Enterprise | Yes | Admin must enable; integrates with Microsoft Agent 365 governance |
| Edu / Healthcare | Beta | Admin opt-in required |
| Plus / Go / Free | No | Not available yet |
| Pro (EU, UK, Switzerland) | No | Still waiting for rollout |
Known gaps at launch: dots can't make phone calls, don't get their own email address, and can't be created on mobile. Launch day itself was rough — a voice demo failed on stage, one dot ignored a live request, and dot.com turned out to belong to xAI (redirecting to the Grokbot download page), while dots.com is a defunct clothing store.
"Included today" doesn't tell you what heavy use costs in November. If you're on Plus, the sensible move is not to upgrade for dots yet — wait for month-one pricing details and for launch bugs to settle.
For context, Meta's Muse agent launched consumer-first in early September and topped both app stores, while dots went premium and work-first. OpenAI's longer-term direction is specialist dots with company identity — it already runs them internally for procurement, invoices, email marketing, support, and contracts — and eventually teams of dots working together.
FAQ
Q: What makes the OpenAI Dots proactive agent different from ChatGPT?
A dot isn't waiting in a chat box. It runs continuously on GPT-6 Astra with its own cloud Linux computer, monitors your connected apps read-only, and surfaces work before you ask. You approve tasks rather than prompt them one at a time.
Q: Is my first dot really free?
For Pro users, the first dot comes at no extra cost, and its usage doesn't count against plan limits during the first month. After that, OpenAI has not announced the terms, and work a dot starts in Codex or ChatGPT already counts toward normal limits.
Q: Can a dot see my passwords or my laptop?
No. Dots never see your passwords, sensitive actions like password changes always stay with you, and access to your local laptop is off by default. An auto review step also checks each action against your rules.
Q: Who can use dots right now?
Pro, Business Premium, Enterprise, and Edu/Healthcare beta users (once an admin enables it). Plus, Go, and free users don't have access, and Pro users in the EU, UK, and Switzerland are still waiting for rollout.