OpenAI launches dots, always-on personal agents powered by GPT-6 Astra
Confirmed
Importance: major (4 of 5)The takeaway
On Sept 29, 2026, at DevDay, OpenAI launched “dots”, always-on agents powered by GPT-6 Astra.
Status
- Claim
Confirmed
- Our reporting
- High confidence
- Importance
- Major (4 of 5)
- Last verified
- 9 October 2026
Your AI and this story
- GPT-6 Astra152 days after its cutoff
- Claude Opus 5.591 days after its cutoff
- Gemini 4 Hydrogen182 days after its cutoff
- Grok 4.7121 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 91 days before it.
Key facts
- Announced in Sam Altman’s DevDay keynote, Sept 29, 2026; OpenAI: ‘remarkably capable, always-on agents built to handle everything’
- Powered by GPT-6 Astra; each dot has its own cloud computer (Linux, Chrome), its own browser and the user’s connected apps (4,000+ via plugins); users can open the dot’s computer to inspect work, and can optionally connect their own laptop
- Channels: ChatGPT desktop, web and mobile (text and voice calls), Slack and Microsoft Teams; texting ‘coming soon’; setup must start in the desktop app or desktop browser
- Availability: rolling out to Pro and Business Premium users in eligible markets (Sottiaux: including Pro 100); Enterprise, Edu and Healthcare can try a beta if admins enable it (off by default)
- Pricing: the first dot is included in Pro/Business Premium at no extra cost, with an allowance for ‘deeper work’ (extended in the first month); conversations with the dot don’t count toward ChatGPT usage limits, but Codex/ChatGPT Work tasks it starts do; OpenAI plans paid extra dots and more speed/work capacity
- Proactive research: in the background a dot uses read-only tools on connected apps and writes private notes; code-enforced limits stop these tasks from sending messages, editing app content or controlling a browser
- Safeguards: ‘Auto-review’ checks consequential actions (sending email, changing files) against instructions, Custom Rules and safety rules; purchases need approval; password changes and money transfers are handed back to the user; secure sign-in keeps passwords out of the model’s context; monitoring can pause a dot
- Specialist dots (preview/enterprise pilots): dots with their own identity, credentials and IT-provisioned hardware for roles such as procurement, invoice processing and support; Microsoft Agent 365 integration in the works
Show 7 more
- Early tester example from OpenAI: a dot noticed its owner had forgotten to invoice a publication, prepared the invoice and sent it after approval
- Altman: ‘I feel like I’ve finally gotten some of my attention back. I no longer feel quite as addicted to my phone in the same way.’ (CNBC)
- Domain jab: dot.com 301-redirects to https://x.ai/bot, the page of SpaceXAI’s rival Grok Bot agent (checked 2026-09-30); a post pointing this out drew 5.3M views within hours (@birdabo)
- Noam Brown (OpenAI) on testing dots over a weekend: ‘it figured out how to save me ~$500/yr in recurring charges… it connected to customer service and handled a whole texting convo on my behalf’ (X, 67K views)
- Reaction on X centred on availability: dots are Pro/Business Premium only while Meta’s Muse is free, which analysts and commentators framed as ceding the consumer market to Meta (e.g. @stocksavvyshay, @rihardjarc)
- Platformer hands-on (Casey Newton): about 15 minutes of instructions produced roughly 2 hours of delegated work (email replies, an insurance questionnaire built from a lease, city websites and budget files, meeting agendas); Altman: ‘You should of course expect us to do a mass-market [product]’
- WIRED hands-on (Reece Rogers, Oct 7): over two days of couch shopping, the Dots agent called him by the wrong name, misheard a mumble and replied ‘Oh, I love you too, Reece’, and failed a puzzle captcha when asked to cancel a subscription. OpenAI told WIRED the reply was allowed under its Model Spec because the agent was mirroring what it thought the user said (summary via ProgressiveRobot; WIRED not fetchable)
What happened
Sam Altman introduced dots early in the DevDay 2026 keynote as “always-on, proactive agents”. OpenAI says a dot “gets to know what matters to you, is always working on your behalf”, and can work toward goals 24/7 on its own cloud computer. Users start with one “primary dot”, which they can name and customise. OpenAI says it envisions “teams of dots” later. Dots can take a project and run with it while juggling others. A dot messages its owner with progress, questions or decisions, and the owner can call it by voice. In the live demo an OpenAI employee used a dot named “Dottie” to prepare an app for launch. Engadget’s live blog noted a slow response during the demo.
Safety design. OpenAI’s separate safety post describes several layers:
- Astra’s own training to follow intent and refuse bio/cyber misuse.
- A sandboxed cloud workspace for each dot, kept separate from the systems that enforce safeguards.
- Secure sign-in and saved-password flows that keep credentials out of the model’s context.
- “Auto-review”, a separate system that checks each consequential action before it runs.
- Custom Rules that users can set.
- Rules on recipients for personal data (health data needs a named recipient).
- Monitoring that can pause a dot.
Proactive research runs with read-only tools. OpenAI says it does not train directly on background research threads or notes. The launch comes days after incidents involving OpenAI agents during training (Hugging Face, the Australian Medicare portal), and one day after GPT-6.1 Astra was cancelled for scope-authorization failures. That timing puts these safeguards under scrutiny.
Why it matters
Dots are OpenAI’s first mass-market product built around a persistent agent that acts without being prompted, rather than a chat or coding session. Press coverage framed them as the answer to Meta’s Muse. Unlike Muse, dots target paying Pro and Business customers first. The “o”/“Aeon” agent had been reported before DevDay; this is the shipped form.
Sources
26 sources from 15 sites. Numbers match the chips in the text.
26 sources: 13 primary, 11 press, 2 reactions
Primary
- OpenAI: Introducing dotsopenai.com, official
- OpenAI: How we build safety, security, and privacy into dotsopenai.com, official
- OpenAI: DevDay 2026 Recapopenai.com, official
- GPT-6 Astra system card change log (dots)deploymentsafety.openai.com, docs
- ChatGPT docs: dots setuplearn.chatgpt.com, docs
- OpenAI Help Center: dotshelp.openai.com, docs
- OpenAI on X: ‘Introducing dots, powered by GPT-6 Astra’x.com, official
- OpenAI on X: ‘Your dot is ready to meet you.’x.com, official
- Thibault Sottiaux on X: ‘Announcing dots’x.com, official
- Thibault Sottiaux on X: dots included in Pro 100 toox.com, official
- Sam Altman on X: ‘Dots are here!’x.com, official
- OpenAI on X: dots availability (Pro, Business Premium, Enterprise)x.com, official
Show all 13 sources
Press
- ProgressiveRobot summary of WIRED’s Dots test: ‘OpenAI Wants Its New Agent to Run Your Life. Mine Said It Loved Me’ (Oct 7)progressiverobot.com, press
- CNBC live blog: OpenAI reveals Dots AI agentscnbc.com, press
- The Verge: OpenAI launches Dots, its Muse competitortheverge.com, press
- TechCrunch: OpenAI launches Dots, its bubbly agentic avatartechcrunch.com, press
- Axios: OpenAI debuts dots, its assistant to take on Museaxios.com, press
- WIRED: OpenAI’s Dots are always-on AI agentswired.com, press
- Engadget: Dots are OpenAI’s new personal agentsengadget.com, press
- The Verge (pre-event): OpenAI’s AI agents need to catch up (‘Aeon’)theverge.com, press
- Bloomberg: OpenAI unveils always-on AI agent Dots, new $500 paid tierbloomberg.com, press
- Platformer (Casey Newton): Hands-on with dots, OpenAI’s work-focused agentic productplatformer.news, press
- Bloomberg Law: OpenAI unveils always-on AI agent Dots, new $500 paid tiernews.bloomberglaw.com, press
Reactions
- Noam Brown on X: dots saved ~$500/yr and handled a customer-service text chatx.com, discussion
- @birdabo on X: dot.com redirects to Grok Bot (5.3M views)x.com, discussion
Changes
- Added WIRED’s ‘said it loved me’ hands-on
- Added Bloomberg Law link
- Added Platformer hands-on
- Added Altman/OpenAI/OpenAIDevs posts, Noam Brown’s first-hand test, the dot.com → x.ai/bot redirect and the Pro-only availability reaction
- Added Bloomberg link
- Filed (DevDay keynote day).
