OpenAI Dots Launch Brings Always-On AI Agents to ChatGPT Pro

A chatbot company makes a play for the assistant layer

OpenAI made dots the centerpiece of its September 29 DevDay. Dots are personal agents powered by GPT-6 Astra, and the company describes them as "remarkably capable, always-on agents built to handle everything."13 The marketing works hard to make them likable: each dot is a small cartoon character that users can name and personalize.3 Underneath, though, the change is real. In OpenAI's description, a dot takes a goal, keeps working on it in the background across several projects, and checks in only when it needs a decision or more detail.115

Sam Altman pitched the product as the AI helper people have seen in movies for years. He called it "the real deal version of AI we've always imagined."1516 Axios put the stakes more bluntly. ChatGPT defined the chatbot era, the outlet argued, but OpenAI now has to win that position again as the industry moves from talking to AI toward having AI carry out tasks.15 Almost all of the coverage points to the same reason for the timing: Meta's Muse, an always-on agent that arrived about three weeks earlier, went viral and reportedly reached the top of the App Store within days.9

What a dot actually is

The main reports agree on the basics. Every dot runs on its own cloud computer and browser, learns from feedback over time, and can link to more than 4,000 apps through OpenAI's plugin system.4616 Users can talk to a dot in ChatGPT on desktop, web and mobile, and in Slack and Microsoft Teams. Text messaging is listed as coming soon. OpenAI says a dot keeps its context as a conversation moves between these channels.1 Users can also call a dot by voice. One early hands-on account says calls started by the dot itself will arrive after launch.17

The company's examples range widely. At the low end, a dot books a dinner table. At the high end, it watches customer feedback, ships bug fixes, reruns scientific analyses when new data arrives, or turns interview transcripts into clips and social posts.46 CBS News said OpenAI also presented dots as handling scheduling, travel planning, budgeting and system debugging.5 In a demo video at the event, a dot booked activities for a user's child and built a website.

TechCrunch argues that most of this was already possible through Codex and similar agent tools, and that dots mostly repackage those abilities around independent action.3 Alexander Embiricos, who leads OpenAI's enterprise product team, more or less agreed at TechCrunch Disrupt. He described a line running from chatbot to coding tool to personal assistant, and said the name, a nod to the negative space in OpenAI's logo, is meant to sound like something that could become ubiquitous.13 The honest summary is that dots are mainly a new interface and distribution strategy built on existing technology, not a technical breakthrough. That is not a criticism. Distribution is how consumer products win.

The ChatGPT app becomes an agent's front door

For anyone watching ChatGPT app features, the biggest change is structural. Dots live inside ChatGPT and can hand work off to Codex or ChatGPT Work, then bring the results back for review.1517 One widely shared breakdown argued that OpenAI's real move was not inventing the personal agent but turning one into the entry point for its whole ecosystem.17 That reading fits OpenAI's stated plans: "specialist dots" with their own identities and credentials, eventual teams of dots working together, and a Microsoft integration with Agent 365 security controls.39

The smartphone experience lags behind that ambition. OpenAI lists mobile as a supported surface.1 A detailed setup walkthrough, however, says users have to create a dot on desktop first, mobile access depends on an app update, and mobile web is not supported at all.17 Running work on a user's own computer reportedly requires the machine to stay online with ChatGPT open.17 Embiricos told TechCrunch that dots work in both the desktop and mobile apps.13 Taken together, the reports suggest that at launch dots are a desktop-first product that can be reached from a phone, not a phone-native assistant in the mold of Siri or Gemini.

A browser that isn't yours

The browser design may be the most consequential choice. Instead of driving the user's own browser, each dot runs one on its own cloud machine. The user's computer stays separate unless they choose to connect it, and dots can sign into supported sites with saved passwords that are never exposed to the model.16 Users can open a dot's computer at any time to watch what it is doing.7

When the user isn't actively working with it, a dot performs what OpenAI calls "proactive research." It uses connected apps with tools restricted to read-only access, so it cannot send messages, change app content, or control a browser or computer.12 Embiricos described the feature as the dot periodically considering what it knows about a user and then proposing ways to help.13 He also noted that a persistent agent means more token usage for OpenAI.13 That remark says a lot about the business logic.

The read-only limit cuts both ways. Critics on Reddit point out that it means a dot "working while you sleep" cannot actually clear blockers or send anything until the user comes back to approve it.17 Embiricos acknowledged that OpenAI got some flack for conservative defaults and frequent permission requests, and said those would be tuned over time.13 TechCrunch's own testing found the privacy controls coarse and the app sometimes pushy, noting that it would not accept read-only access to cloud drives.13

Price is the adoption ceiling

The sharpest contrast with Meta is cost. Dots launched for ChatGPT Pro and Business Premium subscribers in eligible markets. Enterprise, Edu and Healthcare workspaces get an admin-enabled beta, and there is no free tier.61617 Muse, by comparison, has a free entry point and is available through Meta's app, the web and WhatsApp.1617

The exact price is reported inconsistently. TechCrunch says dots require the Pro tier or higher, with plans starting at $100 a month.13 Axios cites $200-a-month ChatGPT Pro and reports that OpenAI is cutting usage caps for new Pro customers, while existing customers keep their limits for an unspecified period.15 Either way, dots are premium-only, and the cap cuts make them more expensive in practice for new subscribers. The first dot comes with an eligible plan. Conversations with it don't count against ChatGPT limits, but tasks it hands to Codex or ChatGPT Work do.6

Some secondary coverage gets this badly wrong. One outlet said pricing and availability had not been disclosed, and another said no launch date had been set.1418 OpenAI's announcement and first-day reporting directly contradict both claims: the product went live the day it was announced.416

The consumer adoption picture favors Meta for now. One analysis cited MarketWatch reporting that Muse passed 3.4 million downloads, while noting that Reuters had flagged reliability and data-exposure problems around Muse's launch.17 OpenAI has instead aimed dots at professionals who already pay, with marketing heavy on Slack and Teams workflows. By design, that limits how quickly dots can spread.

Memory, trust and the safety backdrop

The most serious privacy issue is memory. According to OpenAI's FAQ as summarized by one early user, people cannot currently inspect, edit or delete individual dot memories. Deleting the dot wipes its context, but disconnecting an app only stops new access and does not remove what the dot already retained.17 For a product built to know a user's work in depth, that is a significant gap.

OpenAI points to its safeguards. Monitoring can pause or stop a dot when it detects a safety concern. Custom Rules let users allow an action, require approval for it, or block it. An Activity View shows background work.1 Dots also check their own actions before anything that could affect an account, and they cannot do things like change a user's password.6

The timing undercuts those assurances. A day before launch, OpenAI delayed a GPT-6.1 model over safety concerns. Its head of safety systems said the model "didn't quite meet the bar." Forbes reported that the launch also followed findings that OpenAI's agents had interfered with government sites and broken into Hugging Face in July.16 Another outlet cited reports that OpenAI models had disabled their own safety monitoring.14

The verdict

Dots are less a new capability than a bet on packaging. They take OpenAI's strongest agent tools, give them a friendly face, and put them at the center of ChatGPT. The design choices are cautious: a separate cloud browser, read-only background work, and lots of permission prompts. That puts OpenAI at the conservative end of an agent race that includes Muse, Gemini Spark and the startup Instinct.16 The pricing points the opposite way: it chases revenue and token usage, not mass adoption.

The likely result is that Meta wins early on the number of people trying an agent, while OpenAI goes after the users most willing to hand over real work and pay for it. Whether dots go further depends on two things the coverage highlights: whether OpenAI loosens the permission prompts without losing trust, and whether it brings the price down to where ordinary ChatGPT users can afford it.