OpenAI Dots: Personal Agents on GPT-6 Astra, Explained
Dots (launched Sep 29, 2026) is OpenAI's entry in the personal agent war: always-on agents, each with a persistent cloud computer and browser, running on GPT-6 "Astra". Unlike Meta Muse or Manus Cue, Dots is not a separate app — it lives inside ChatGPT across app, desktop and web.
Verified facts (public information)
- Persistent cloud computer per instanceEach Dot gets a dedicated cloud machine and browser. Tasks survive locally — this is a persistent runtime, not a temporary sandbox.
- GPT-6 "Astra" under the hoodDots run on OpenAI's GPT-6 Astra model generation, positioned as the reasoning engine for always-on autonomy.
- Read-only restrictions until approvedUnapproved Dots are limited to read-only background research. Write actions (purchases, emails, bookings) unlock after approval gates.
- Memory persists across runs — no fine-grained delete UIYour Dot accumulates context across sessions. There is currently no per-item memory deletion interface; memory pruning is not user-controllable yet.
- Embedded, not a separate appUnlike Meta Muse (standalone app) or Manus Cue (standalone app), Dots is embedded across ChatGPT surfaces: app, desktop and web.
Facts above are from OpenAI's public launch communications and press coverage. Unknowns (unit economics for non-Pro tiers, memory-pruning internals, GA timing) are intentionally omitted.
Access: who gets Dots, and when
Dots are rolling out progressively — high-tier accounts first(Pro-level memberships), appearing directly inside ChatGPT surfaces when eligible. There is no public waitlist form yet. If you don't see Dots in your ChatGPT interface, your account is not in the current rollout cohort.
Dots vs Muse vs Cue — the one-paragraph version
Dots: persistent cloud computers embedded in ChatGPT; read-only until approved; memory persists. Muse: standalone app; isolated Linux VMs; eBPF-monitored egress; virtual payment cards; consumer-life focus (booking, bills). Cue: invite-only client-side planner with per-agent email/phone/wallet/virtual computer; lightest and most consumer-ready of the gated launches. Full comparison: Personal Agents battle guide.
FAQ
What are OpenAI Dots?
Dots are OpenAI's always-on personal AI agents, launched Sep 29, 2026 on the GPT-6 "Astra" model generation. Each Dot is an autonomous agent with its own persistent cloud computer and browser that works toward goals without prompting — OpenAI's direct answer to Meta Muse.
How do I get access to OpenAI Dots?
Dots are rolling out gradually, with high-tier accounts (Pro-level memberships) getting access first. There is no public waitlist signup yet — access appears inside your ChatGPT app, desktop and web interfaces when your account is eligible. OpenAI has not announced general availability timing.
How is Dots different from Meta Muse?
Dots is embedded inside ChatGPT (no separate app) and gives each instance a persistent cloud computer. Muse is a standalone Meta app that runs tasks inside isolated Linux VMs with virtual payment cards. Muse targets everyday life (booking, bills, shopping); Dots initially targets research and task execution with read-only restrictions until approved.
Is my Dots memory private? Can I delete it?
Dots retain memory across runs and there is currently no fine-grained deletion interface — you cannot selectively remove individual memories yet. Treat anything you tell your Dot as long-lived context, and avoid connecting credentials you would not want retained.
More ecosystems: Personal Agents battle guide · TypeSafe JEV (System One models) · JEV plugin directory