# Meta's Muse agent: what it is and what it means for businesses

> Muse is Meta's personal AI agent that sends emails, fills out forms and negotiates for the user. See what it does, where it works and how to prepare support for customers who send an agent.

- **HTML version:** https://tolky.to/en/blog/meta-muse-personal-ai-agent
- **Author:** [Marlos Carmo](https://tolky.to/en/author/marlos), Co-founder of Tolky
- **Published:** November 19, 2026
- **Reviewed:** November 19, 2026
- **Language:** en

## Summary

The **Muse agent** is Meta's personal AI agent, announced on September 8, 2026: it runs long tasks for the user, such as sending emails, booking travel, filling out forms and negotiating, and it can be used in the Muse app or on WhatsApp. As of November 19, 2026, it works in the United States, Singapore and Australia; Meta has not announced other countries, such as Brazil, in the pages we checked. For businesses, the message is to get ready for customers who reach out through an agent: clear public information, confirmation before sensitive actions and handoff to people.

Until 2025, whoever contacted your business on WhatsApp or by email was a person. In 2026, a third kind of counterpart starts to show up: the customer's AI agent, which asks, compares, fills out and negotiates on the customer's behalf.

Meta's Muse agent is the most visible example of this shift. This guide explains what it does, where it is already available, what Meta promises about privacy and what a support operation should adjust before these agents reach more countries.

*Updated in November 2026, with Meta's official pages checked on November 19, 2026.*

## What is Meta's Muse agent?

**The Muse agent is Meta's personal AI agent, announced on September 8, 2026, which runs long tasks on the user's behalf instead of only answering questions.** In Meta's words, it "doesn't just answer questions, it does the work" ([Meta's announcement](https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/), September 8, 2026, updated November 10).

The agent runs on Muse Spark, the main model in the [Muse family](https://tolky.to/en/blog/what-is-meta-muse). The user talks to it in the Muse app or directly on WhatsApp, and the agent keeps working in the background until the task is done.

It is different from [Meta AI](https://tolky.to/en/blog/meta-ai-on-whatsapp), the free assistant that answers questions on WhatsApp, and from Meta Business Agent, which serves customers on behalf of businesses. We compare the three in [which Meta AI to use in your business](https://tolky.to/en/blog/meta-ai-for-business).

## What can the Muse agent do?

**According to Meta, the Muse agent sends emails, books travel, fills out forms, negotiates and follows tasks that take hours or days, asking for confirmation before sending messages or making purchases.** This is what the market calls [agentic AI](https://tolky.to/en/glossary/agentic-ai): AI that plans steps and uses tools to reach a goal.

- **Communication:** writes and sends emails after your approval.
- **Bookings and purchases:** researches options, books travel and completes purchases with confirmation.
- **Forms:** fills out sign-ups and requests on websites.
- **Negotiation:** talks to other parties to get better terms.
- **Long tasks:** keeps working in the background and lets you know when it is done.

At the Connect event, on September 23, 2026, Meta announced Mac control, a dedicated email address for the agent, support for its glasses and commerce partnerships ([TechCrunch](https://techcrunch.com/2026/09/23/everything-new-coming-to-metas-ai-agent-muse/)).

## Where is the Muse agent available?

**As of November 19, 2026, the Muse agent works in the United States (since September 8), Singapore (since October 27) and Australia (since November 17); Meta has not announced other countries, such as Brazil, in the pages we checked.** The sources are Meta's announcements for [Singapore](https://about.fb.com/news/2026/10/muse-is-now-available-in-singapore/) and [Australia](https://about.fb.com/news/2026/11/muse-is-now-available-in-australia/).

| Item | What Meta published | Source |
| :--- | :--- | :--- |
| Launch | September 8, 2026, in the United States, on iOS, Android and the web | September 8 announcement |
| Expansion | Singapore (October 27) and Australia (November 17) | Country announcements |
| Other countries | Not mentioned in the pages we checked | Checked on November 19, 2026 |
| Price | Free for most use, with subscription plans | September 8 announcement |
| Model | Muse Spark | September 8 announcement |
| Where to chat | Muse app or WhatsApp | September 8 announcement |

Meta has not published subscription prices on the pages we checked. The prices circulating on third-party sites have not been confirmed by Meta, so we do not repeat them here.

## How does the Muse agent protect user data?

**The Muse agent runs on a dedicated virtual machine in the cloud, the Muse Secure VM, which stores the user's data and credentials; Meta says it has no access to passwords or payment methods.** A control agent called Sentinel approves each internet access.

The promises Meta has published:

- **Confirmation** before sending an email or buying.
- **Activity log** so the user can check what the agent did.
- **Opt-out** from having the data used to train models.
- **Separation from advertising:** the agent's data does not go to ad systems.
- **Muse Confidential VM,** announced as the next step, with keys held only by the user.

A promise is not a guarantee. In October 2026, Meta paused Muse conversation share links after about 6,800 of them showed up in public records, according to Ars Technica reporting picked up by [PPC Land](https://ppc.land/meta-halts-muse-share-links-after-6-800-turn-up-in-public-dns-records/) on October 21. For businesses, the lesson is the same as with any [AI platform](https://tolky.to/en/blog/security-data-privacy-enterprise-ai-platforms): check what is promised, what can be audited and what stays with you.

## Will the Muse agent talk to my business?

**It already does, indirectly: the agent sends emails, fills out forms and negotiates with businesses on the user's behalf, and the US press reported a test of phone calls to businesses in the United States.** Meta does not say, in the pages we checked, that the agent sends messages to businesses on WhatsApp.

According to [404 Media](https://www.404media.co/meta-tests-muse-ai-agent-calls-that-are-actually-made-by-humans-in-a-call-center/), on September 22, 2026, the agent was making test calls to US businesses, some of them completed by human call center agents. Meta replied that it will only launch the feature when it is ready and with the proper disclosures. A day earlier, [SiliconANGLE](https://siliconangle.com/2026/09/21/amazon-blocks-metas-muse-agent-from-shopping-on-users-behalf/) reported that Amazon blocked Muse from shopping on users' behalf.

Both episodes point to the question every business will have to answer: when the customer sends an agent instead of coming in person, what will you agree to do without talking to the person?

## How do you prepare customer service for customers' AI agents?

**Prepare your customer service with four measures: clear and consistent public information, confirmation from the account holder before sensitive actions, clearly identified official channels and handoff to a person when in doubt.** None of this depends on Muse reaching your market; it applies to any agent.

| Situation | Risk | What to do |
| :--- | :--- | :--- |
| Agent asks about price, timing or policy | Different answers on each channel | Publish the same information on the website and in the support [knowledge base](https://tolky.to/en/glossary/knowledge-base) |
| Agent requests a cancellation, exchange or refund | Action without the customer's real intent | Confirm with the account holder through a channel already on file |
| Agent requests third parties' personal data | Data leak and a breach of Brazil's data protection law ([LGPD](https://tolky.to/en/glossary/lgpd-in-customer-service)) | Only release it to the identified account holder, never to the agent |
| Agent negotiates a discount | Concession outside the policy | Written limits in the support agent and approval authority for people |
| Agent insists or repeats requests | Busy queue and cost per conversation | [Escalation](https://tolky.to/en/glossary/escalation) rules and a limit on attempts |
| Agent called by voice | Customer does not know what was agreed | Written summary sent to the contact on file |

Three principles help you write these rules:

1. **The customer's agent is an intermediary, not the account holder.** Actions involving money, contracts or personal data require confirmation from the customer.
2. **Your AI needs limits too.** A support [AI agent](https://tolky.to/en/glossary/ai-agent) should know what it can grant and when to call someone; see the [rules for handoff from AI to humans](https://tolky.to/en/blog/ai-to-human-handoff-7-rules).
3. **Everything is logged.** Who asked, through which channel, what was done and who approved it. This is [AI governance in customer service](https://tolky.to/en/blog/ai-governance-in-customer-service).

## What changes for sales and marketing?

**When customers hand research off to an agent, the winners are businesses with clear, structured and easy-to-verify information; vague promises and hidden pricing lose ground.** Agents compare quickly and rarely second-guess objective data.

- **Public pricing and terms** cut down on back and forth. A readable pricing page, like [our pricing page](https://tolky.to/en/pricing), serves both people and machines.
- **Consistent answers** across website, WhatsApp and email keep the agent from finding contradictions.
- **Response time** matters more: an agent that waits for hours may simply look for another business. See the [hidden cost of slow customer service](https://tolky.to/en/blog/hidden-cost-of-slow-response-time).
- **Relationships stay human:** the agent brings the request, but trust and repeat purchases still depend on how the business treats the customer.

## Where does Tolky fit in?

**Tolky is a conversational AI CRM: its AI agents serve customers on WhatsApp through the official API, on the website, by email, on Microsoft Teams and by voice, with limits written by the business, handoff to people with the history and a record of every conversation in the CRM.** This applies both to the customer who writes and to the agent who speaks for them.

![Tolky's Contacts screen in table view, with period filters and, for each contact, name, email, phone, creation date and last interaction, plus search and export](https://tolky.to/screenshots/contatos-crm-en.png)

- **Rules per mission:** what the agent can answer and grant, and when it hands off the conversation. See the [AI chatbot for WhatsApp](https://tolky.to/en/features/whatsapp-chatbot).
- **Customer records:** each contact has a history and known channels, which helps confirm requests with the account holder. See [Tolky's CRM](https://tolky.to/en/solutions/crm).
- **Activity log** with export, to audit what the AI and the team did.

**What Tolky does not do:** it does not identify on its own whether the sender is a person or an AI agent. Protection comes from the confirmation and handoff rules you configure.

## Frequently asked questions

### What is Meta's Muse agent?

It is Meta's personal AI agent, announced on September 8, 2026. It runs tasks for the user, such as sending emails, booking travel, filling out forms and negotiating, and it runs on the Muse Spark model.

### Is the Muse agent available outside the US?

Yes, in two more countries as of November 19, 2026. Meta launched the agent in the United States and brought it to Singapore and Australia, without announcing dates for other countries, such as Brazil, in the pages we checked.

### Is the Muse agent free?

Meta says it is free for most of what people need and that there are subscription plans. Plan prices do not appear on the official pages we checked.

### What is the difference between the Muse agent and Meta AI?

Meta AI is an assistant that answers questions, summarizes messages and creates images inside Meta's apps. The Muse agent goes further: it runs multi-step tasks on the user's behalf, with confirmation before sending or buying.

### Can the Muse agent talk to businesses on WhatsApp?

Meta says users can chat with Muse on WhatsApp, but it does not state, in the pages we checked, that the agent sends messages to businesses there. It already acts with businesses through email, forms and, in a test in the United States, by phone.

### How should my business handle a request made by an AI agent?

Answer public questions as usual, but confirm with the account holder before canceling, refunding, changing a contract or releasing personal data. Log who asked and through which channel, and hand off to a person when in doubt.

---

## Conclusion

The Muse agent works in only three countries so far, but it shows where customer service is heading: part of the requests will come from agents speaking on customers' behalf. A business that publishes clear information, confirms sensitive actions with the account holder and has handoff rules serves both people and agents well.

[See the plans](https://tolky.to/en/pricing), [learn what Muse is](https://tolky.to/en/blog/what-is-meta-muse), [read the AI agent guide](https://tolky.to/en/blog/what-is-an-ai-agent-complete-guide) or [talk to the team →](https://tolky.to/en/contact). Let's talk?

---

This is the Markdown version of the page. The full version, with images and navigation, is at https://tolky.to/en/blog/meta-muse-personal-ai-agent

Tolky: https://tolky.to
