OpenAI Dots Are Always-On Agents: What Changes
Dots launched at DevDay Sep 29 on Astra with own cloud box and 4,000 apps. How primary and specialist Dots work, what they cost, and rules before Slack access.
On this page
OpenAI launched Dots at DevDay on September 29, and the pitch is simple: name one agent, give it goals, and let it work while you do things. Powered by GPT-6 Astra with its own cloud computer and over 4,000 apps, Dots live in ChatGPT, Slack, and Teams. Here is what they do, what they cost, and how to scope them.
Takeaways
What shipped: primary Dot you name and shape, plus specialist Dots preview with identity and IT controls. Where it runs: own cloud computer and browser, context carries across ChatGPT, Slack, and Teams. Entry: Pro and Business Premium in eligible markets, Enterprise pilot on admin enable, first Dot included. Builder note: Agents API adds computer use, Codex multi agent handoff, tool search, and compaction, so Dots sit atop plumbing you can also call directly.
What can a Dot actually do?
Give a Dot a standing goal and it keeps working without a chat open: watch a Slack channel for bug alerts and dig in, track an invoice and submit after your approval, or rerun analysis when new data lands. OpenAI frames examples by role: a developer Dot monitors feedback and opens fixes, a scientist Dot rechecks results, a sales Dot preps briefs. Conversations with your Dot do not count toward ChatGPT limits, while tasks it starts in Codex or ChatGPT Work count as usual.
start: name one Dot in ChatGPT, connect apps, state the standing goal
runs: own cloud computer plus browser across 4,000 plus apps
reach: message in ChatGPT, Slack, Teams, voice calls, texting soon
team: single Dot now, teams of Dots cooperating later per OpenAIPricing and limits to note
New Pro 500 at $500 per month offers 25 times Plus allowance plus Ultrafast mode, with Pro 200 reopened under new allowances. OpenAI counts 1.2 billion weekly ChatGPT users after crossing 1 billion in July. Sensitive acts like password changes stay with you, and OpenAI notes Dots can still make mistakes, so review consequential work.
How do you set one up safely?
Start with one narrow goal
Pick a single loop like triage support pings or prep a daily brief. Broad mandates like handle my inbox create the exact failure mode personal agents keep hitting.
Connect minimum apps first
Link one source plus one action target, not your whole drive and mail on day one. Add read first, write after a week of clean runs.
Require approval on spends and sends
Keep invoice submit, purchase, and external message steps on human confirm. The invoice example OpenAI cites only works because the tester approved before submit.
Plan for Agent 365 if you scale
OpenAI is working with Microsoft to manage specialist Dots through Agent 365 security controls. Name owners, credentials, and kill paths now so IT can adopt that control plane later.
What about GPT-6.1 Sol and the Agents API?
DevDay paired the agent wrapper with cheaper muscle. GPT-6.1 Sol targets near Astra skill for coding, computer use, and professional work at one fifth Astra token price. Ultrafast promises up to 8 times faster token output in Codex and 6 times in the API, with Astra now and Sol soon. The Agents API in public beta adds hosted runs, memory, multi agent delegation, plus computer use so agents operate software through its UI, and a Decisions API preview uses Luna to classify and route.
Should you hand Slack to a Dot today?
How are Dots different from Meta Muse?
Muse targets free consumers and topped app charts after its September 8 launch, while Dots target paid Pro, Business Premium, and Enterprise work use. Both use cute avatars and background work, but Dots lean to org controls and specialist identity where Muse leans to everyday errands.
What should EU and UK users expect?
Community reports note Dots rolling to Pro users except the European Economic Area, Switzerland, and the UK at launch. If you build for those regions, keep a Codex plus Agents API path until availability widens, and pin model IDs per market.
As of October 1, 2026: Dots move agents from sessions you open to goals that run while you are away, but scoping still decides safety. Start narrow, approve spends, and log every action. Next, read agents that do work for the harness pattern Dots now productize.