Muse review, pricing and verdict

Muse suits a US or Canadian small business owner who already sells and advertises on Facebook and Instagram and wants an agent that reads the shop's sales, books and ad results, then drafts the next move for approval. It is a weak pick for anyone outside those two countries or anyone who needs published usage limits before committing.

Muse is Meta's personal AI agent, launched 8 September 2026, that works toward a goal inside its own cloud computer and pauses for approval before it posts, sends or spends. Since 29 September its Small Business skills connect Instagram, Facebook Pages and Meta ad accounts plus tools like Shopify, QuickBooks and Slack.

Maker: Meta · Autonomy: semi autonomous · Maturity: GA

Underlying models: Muse Spark

About Muse

Muse is the personal AI agent from Meta, built by Meta Superintelligence Labs under chief AI officer Alexandr Wang and launched on 8 September 2026. Three weeks later, on 29 September, Meta added Muse for Small Business: not a separate app but a set of business skills and connectors inside the same agent, announced a day after Meta set up its Meta Enterprise Platform with former MongoDB chief executive CJ Desai in charge. Muse runs on Muse Spark, Meta's own proprietary model, rather than on the open-weights Llama line.

The agent takes a goal instead of a prompt. Each user's Muse works in a private cloud machine Meta calls the Muse Secure VM, where the person can watch its browser and step in, and the work carries on once the app is shut. A separate system called Sentinel decides whether each action on the internet or a connected service is allowed, blocked or sent to the person for approval, and every step lands in an audit trail. People reach it through the iOS, Android, Mac and web apps or by messaging it inside WhatsApp.

For a business owner the change is context. Muse can pull in a shop's Instagram insights, its Facebook Page and its Meta advertising together, then read the tools the shop already runs on: Meta names Canva, Figma, Notion, Granola, Lovable, Shopify, Slack, Stripe, Intuit QuickBooks, Klaviyo, HighLevel, Asana, Box, Dropbox and Zoom, plus custom connectors for anything missing. Meta's suggested jobs are a yearly growth plan from sales and campaign data, a triage of email and calendar, a draft ad campaign built on what is trending, and a monthly check for expenses that look off. Nothing publishes, sends or spends without the owner's approval.

Where Muse is weaker is proof and reach. Access is limited to two countries, the US and Canada, Meta publishes no usage numbers for any plan, and the version Meta itself cannot see inside, a confidential VM, was still promised rather than shipped when HokAI checked on 19 October 2026. Model queries can be used by Meta for training unless the person turns that off, although Meta says Muse conversations are not passed to its ad systems.

Who should look elsewhere: a startup that wants an agent to run outbound, paid search and SEO pages end to end is closer to Revnu, and a team that already keeps its pipeline in HubSpot gets more from that system's own automation than from a second agent reading it. Someone who mainly wants research, writing and coding help across many countries is better served by ChatGPT or Claude, developers wanting a browser agent behind an API fit MultiOn, and rule-based app-to-app automation is still Zapier's job. Our Grok vs Muse vs OpenAI Dots comparison covers the consumer side, and more options sit in the AI agents directory.

Pricing

Meta sells one Muse with a free tier and two subscriptions, $20 and $100 a month, as Axios reported at the 8 September launch; Alexandr Wang told Axios most people should be able to stay on free. Muse for Small Business sits inside those same plans, and Meta's business page says Muse is free for most of what people need, with subscription plans for those who want more. Meta's subscriptions help page names them: Power at $20 a month with 500M Muse tokens a week, and Maximum at $100 a month with 3B a week. It does not publish the free allowance, so the in-app plan screen is the place to check that. For a growth agent with a published business list price, compare [Revnu](/hub/tools/revnu).

Key Features

• Goal-based task execution: Hand Muse a goal such as running a promotion or finding new customers and it plans the steps, works through them in its own browser, and keeps going after you close the app.
• Muse Secure VM: Each person gets a persistent, isolated Linux computer in Meta's cloud with a full browser they can watch and take over at any point.
• Sentinel approval gate: A separate Sentinel system rules on every action, allowing it, blocking it or sending it to you, and permissions can be scoped to 1 task, service, transaction or time window.
• Small business connectors: Links Instagram analytics, Facebook Pages and Meta ad accounts in one step and reads 15 named business tools including Shopify, QuickBooks, Klaviyo and HighLevel, plus custom connectors.
• WhatsApp and Mac access: Message the agent inside WhatsApp the way you would text a person, or use the iOS, Android, Mac and web apps.
• One-time card numbers at checkout: When Muse buys something it generates a single-use card number, so neither the merchant nor the agent sees your real card, and eligible purchases carry Link's purchase protections.

Strengths

• Free for most use, according to Meta, so a shop owner can test it on real sales and ad data before paying anything.
• Meta ad accounts, Facebook Pages and Instagram analytics connect in one step, which no third-party agent can match inside Meta's own apps.
• Approval-first by design: Sentinel gates each action, nothing posts or spends without sign-off, and the audit trail shows what it did and plans to do.
• Saved logins sit in a credential store the agent cannot read, and purchases use one-time card numbers.

Weaknesses

• Available only to adults in the US and Canada, with no date for other markets.
• Meta publishes no free-tier allowance and does not say what the $20 and $100 plans add, so cost at heavy use is guesswork until you hit a limit.
• Model queries can be used for training by default and must be switched off, and the confidential VM Meta promised for before year end had not shipped by 19 October 2026.
• It reads business data but is not a system of record: there is no built-in CRM, books or storefront, so it depends on the tools it connects to.
• Still a weeks-old product with no independent reviews of its business skills, so results rest on Meta's own examples.

Visit Muse Official Site