Instinct vs Meta Muse vs ChatGPT Dots: Personal AI Agents Compared

Instinct, Meta Muse and ChatGPT Dots are all personal AI agents that act for you across email, calendars and online services, but they reach you in different ways. Instinct lives in your text messages and is invite-only; Muse is a Meta app that also runs inside WhatsApp and is open to US users; Dots sit inside ChatGPT and are limited to the Pro and Business Premium plans. The biggest differences are who can get in and how each one handles permissions.

All three arrived within eight months of each other, and two of them within the same September. The facts below come from the three companies’ own pages and from TechCrunch, Engadget, SiliconANGLE and Fortune, as of October 2026. We have not tested any of the three. Where a figure comes from a single outlet or sources disagree, we say so.

The three agents in one table

The last row is a list of claims. None of the three companies has published an independent security audit of its agent.

Instinct: the texting assistant from a startup

Instinct is the oldest of the three and the smallest company. Founder Noah Shinn, a former Sierra research scientist, built it around one idea: you should not need a new interface. You text or call it, and it works through your connected accounts. The privacy policy lists Google Workspace connections (Gmail, Calendar, Drive, Docs, Sheets, Slides, Tasks) and allows you to give it third-party usernames and passwords.

It has also gone furthest towards acting as a person in its own right. It has its own email addresses for contacting businesses (TechCrunch), it can place calls through Concierge, and since 5 October it can join group chats, even with friends who have no account (TechCrunch).

On permissions, Instinct’s position is set out mostly in its terms and in statements by Shinn. The terms appoint Instinct as your agent and say any agreement it enters “shall be binding on you as if entered into directly by you”. Confirmation steps are possible but not promised. After a disputed incident in September, Shinn described isolated sandboxes, short-lived local credentials and identity-signed tool execution. For group chats he says your agent asks before connecting to a group, before sharing information and before taking action.

Access is the main obstacle: you need an invite. A full profile is in What is Instinct AI?.

Meta Muse: the mass-market option

Muse launched on 8 September 2026 and runs on Muse Spark, which Meta calls its most capable model to date (Meta newsroom). It can send emails, book travel, fill in forms, negotiate and lower bills, make purchases and keep working after you close the app. It also makes suggestions without being asked.

Scale is its advantage. According to Apptopia estimates cited by Fortune, Muse reached about 2.8 million installs worldwide in its first 12 days and replaced ChatGPT as the top free app in the US App Store. Meta has since added a Mac app that works with files, messages, calendar, notes and mail, with opt-in access, and launched Muse for Small Business with integrations such as Shopify, Slack, Notion, QuickBooks and Stripe.

On availability, Meta and TechCrunch say the US; Fox News says the US and Canada, and TechCrunch later reported Muse topping charts in both countries. Meta describes access as free for most of what people need, with subscription plans for more. TechCrunch reports a free tier plus two paid plans, Power at $20 a month and Maximum at $100 a month (as of October 2026), and a payment card requirement at sign-up, with a usage meter that warns when free usage runs out.

Muse has the most detailed published security design of the three:

• the agent and your data run in Muse Secure VM, a dedicated cloud virtual machine other users’ agents cannot reach
• a separate agent called Sentinel, isolated at system level, must approve anything Muse sends to the internet
• Muse cannot see passwords or payment methods; payments go through one-time Link by Stripe cards
• apps are connected one at a time, and for email you choose read-only or read-and-send
• sensitive actions such as sending email or buying need your approval, and there is an audit trail

Meta says Muse conversations are not shared with its ad systems. You can opt out of your interactions being used for training, which CNBC reported means training is on by default. TechCrunch noted that these claims “will require deeper investigation by security experts”, and Meta’s privacy record, including the 2019 FTC penalty, is the context most readers will bring to it.

ChatGPT Dots: the agent inside a subscription

OpenAI announced Dots at DevDay on 29 September 2026, running on GPT-6 Astra. Each dot gets its own cloud computer and browser, works in the background around the clock and connects to more than 4,000 apps through plugins (Engadget). You can open the dot’s computer and watch what it is doing. Connecting a dot to your own computer is possible but off by default.

You message or call your dot inside ChatGPT, or reach it in Slack and Microsoft Teams. Texting is announced as coming soon. That puts Dots closer to work tools than Instinct, which is built around personal messaging.

Access is the narrowest for individuals. Dots are rolling out to ChatGPT Pro and Business Premium users in eligible markets, and Pro is not offered for Dots in the EEA, Switzerland or the UK. Enterprise, Edu and Healthcare customers get a beta that an admin must switch on. The first dot is included with the plan, and its usage does not count against plan allowances during the first month; extra dots are planned at prices OpenAI has not disclosed. Reported plan prices differ between outlets and we could not confirm them on OpenAI’s own site, so check the current figures on OpenAI’s pricing page.

The control model is rule-based. Custom Rules let you set what a dot does on its own, what needs your approval and what it never does. An auto-review step checks actions that touch your accounts or share information, a monitoring system can pause or stop a dot, and during “proactive research” app access is read-only. Saved passwords are used without exposing them to the model. For businesses there are specialist dots with their own identity and credentials and integration with Microsoft Agent 365 for governance.

What is missing is a primary source on privacy and training for Dots. OpenAI’s own help pages were not accessible to us, and we do not repeat secondary summaries here.

Which personal AI agent should you choose?

For most people the choice is made by access, not features. In the UK or EU, Muse is US-focused, Dots are not offered on Pro, and Instinct has no country list and no GDPR reference in its privacy policy. For many European readers, the realistic option today is to wait.

If you can choose, our view is:

• Muse if you want granular, documented permissions and are comfortable with a Meta account. It is the only one where read-only email access and one-time payment cards are part of the published design.
• Dots if your work already runs on ChatGPT, Slack or Teams, and your company wants admin control over what an agent may do.
• Instinct if you value texting and calling above everything else and accept a startup’s terms, which put responsibility for the agent’s actions on you. The safety record is covered in Is Instinct AI safe?.

Competition between them is open. Meta’s Alexandr Wang mocked Instinct on X as “Insect”, Fortune reported, and OpenAI announced Dots three weeks after Muse went live.

What companies should take from the comparison

The three products agree on more than their marketing suggests. All three connect to email and calendars, all three buy things, and all three rely on some form of approval step for risky actions. They differ in whether that approval is guaranteed, configurable or merely possible. Muse promises it for sensitive actions; Dots let you set it; Instinct’s terms allow it without promising it.

That is the same design question any company faces when it builds its own agent. Which tools can it call, with which credentials, and which actions need a human click? A decision layer that answers those questions before the model acts is the subject of our piece on Jev AI. For agents built on your own data and systems, see our generative AI integration services and our overview of AI agent companies.

FAQ

Is Meta Muse free?

Muse has a free tier and two paid plans, Power at $20 a month and Maximum at $100 a month as of October 2026, according to TechCrunch, which also reports that you need to add a payment card to start. Meta describes it as free for most of what people need. Current plans are on the Muse site.

Are ChatGPT Dots available in the UK or Europe?

Not on ChatGPT Pro. Dots are rolling out to Pro and Business Premium in eligible markets, but Pro is excluded in the EEA, Switzerland and the UK. Enterprise, Edu and Healthcare workspaces can get a beta if an admin enables it.

Which is safer, Instinct or Muse?

Neither has been independently audited. Muse has the more detailed published design, with a separate approval agent, read-only email options and one-time payment cards, while Instinct’s terms do not guarantee confirmation before actions.

Can Muse or Dots join group chats like Instinct?

TechCrunch reports that Muse has no group chat function, although Meta AI can be added to WhatsApp groups. We found no report of Dots joining group chats.

Which personal agent can make phone calls?

Instinct Concierge places calls in early access. Muse calls to US businesses are reportedly in limited testing. For Dots, you can call your dot inside ChatGPT, but we found no report of it calling businesses.

Building an agent for your own company

Consumer agents are setting expectations for what software should do without being asked. Companies that want the same convenience for staff or customers need the controls the consumer products are still working out: scoped access, approval rules and an audit trail. AI Superior designs and builds agents of this kind for businesses, starting with which actions the agent may take alone and which must wait for a person.