Meta brings its Muse AI agent to small businesses: what it connects to and what it can do

Meta has released Muse for Small Business, a version of its Muse AI agent that connects to the accounting, payments, store and messaging tools many small companies already run on. It matters because Muse is built to act rather than only answer: it can draft campaigns, read financial records and prepare emails across those systems, which puts real business data and money within reach of an AI agent from one of the largest advertising companies in the world.

What happened

Meta announced Muse for Small Business on 29 September 2026. The post opens by describing Muse as "a personal AI agent available in the US and Canada that completes tasks on your behalf". The consumer version had been introduced three weeks earlier, on 8 September, when Meta said it was launching in the United States on iOS, Android and the muse.ai website, with support for its AI glasses to follow.

The small business version is built around goals rather than single prompts. An owner describes what they want, and Muse plans and carries out the steps. Meta says it can also work proactively, for example by flagging emails that need a reply and drafting responses, and that an "Ideas" tab suggests further tasks. The examples Meta gives include drafting an annual growth plan, sorting priorities across email and calendar, improving advertising campaigns and reviewing monthly financial performance.

Meta lists connectors for the following services at launch:

• Asana, Notion and Slack for projects and team communication
• Box and Dropbox for files
• Canva, Figma and Lovable for design and building
• Intuit QuickBooks and Stripe for accounting and payments
• Shopify, Klaviyo and HighLevel for selling and marketing
• Granola and Zoom for meetings
• Facebook Pages, Instagram business accounts and Meta ad accounts

Custom connectors are also supported. Meta says that "in the months ahead" it expects to add more connectors and skills.

The launch followed a broader announcement on 28 September, when Meta introduced Meta Enterprise Platform, an effort to sell its models, agents and developer tools to businesses. Meta said the platform would initially bring the Muse agent, Meta Business Agent, the Muse API and Muse Code to businesses and developers, and that it had hired a former MongoDB chief executive to lead it.

How it works

Meta says Muse runs on a model it calls Muse Spark, which it describes as built for "real-world agentic work". The more important detail for a business is how the agent is contained. In a technical post published alongside the consumer launch on 8 September, Meta's research team set out the design:

• Each user's agent runs in a dedicated cloud virtual machine, isolated from other users' agents.
• A separate component called Sentinel decides what the agent may do. Meta writes that "Sentinel is the sole permission authority for approval to perform actions with connectors to third-party services and for all egress", meaning all traffic leaving the machine.
• The model never handles real login tokens. According to Meta, "the agent never sees real tokens", so a malicious instruction hidden in a web page or email cannot trick it into revealing them.
• When an action needs a person's consent, "Sentinel creates a pending approval and execution stops" until the user responds.
• Connectors separate read access from write access, so a user can, in Meta's words, have "fine-grained control over exactly which actions the agent can take".

Meta also says it takes a "defense-in-depth" approach to prompt injection, combining model training, protections in the software around the model and classifiers. Conversations are sanitised before being used to train models, and users can switch that use off in settings.

For small businesses, Meta's headline promise is simple: "nothing publishes, sends, or spends without your approval." Its consumer announcement added that Muse shows "a complete audit trail of everything it has done and plans to do".

On cost, Meta says Muse "is free for most of what people need" and that "for those who want more, we're offering subscription plans". TechCrunch reports that the free tier carries usage limits and that paid plans raise them. Meta has not published prices for those plans in the material we reviewed.

Meta is not alone here. TechCrunch names OpenAI and Google as the main rivals in agents that act on a user's behalf. Some of the tools Muse connects to have their own built-in assistants: Shopify, for example, says its Sidekick agent "is included with your Shopify plan", with features and usage limits that vary by plan.

Why it matters

Most small firms do not have staff to stitch software together, which is the gap Meta is aiming at. TechCrunch reports that Meta framed the product around owners who are "short on hours, not ideas". An agent that can read last month's QuickBooks figures, look at Shopify sales and draft an Instagram campaign in one conversation could save real time if it works as described.

The same breadth is the risk. Connecting an agent to accounting, payment and customer systems concentrates access in one place. Meta's published design addresses the obvious concerns, such as credential theft and unapproved actions, but those safeguards are Meta's own description and have not, as far as we have seen, been independently audited. Meta says a future "Muse Confidential VM" will allow cryptographic verification that Meta cannot access user data; that is not available yet.

There is also a commercial angle worth keeping in view. Meta earns most of its revenue from advertising, and Muse connects directly to Meta ad accounts and can draft and improve campaigns. Owners should judge any spending suggestions on their own merits.

What we do not know yet

• Prices. Meta has not published what the subscription plans cost or exactly where the free tier's limits sit.
• Countries. Meta's post describes Muse as available in the US and Canada. It has not said when, or whether, the small business version will reach the UK, the EU or elsewhere.
• Connector permissions in practice. Meta describes fine-grained read and write controls, but the exact scopes each third-party connector requests are not set out in the announcement.
• Data handling for business records. Meta explains training opt-out for conversations, but it has not detailed retention periods for data pulled from connected business systems in the material we read.
• Accuracy. There is no independent evidence yet on how reliably Muse handles financial analysis or multi-step tasks.

What it means for you

If you are in the US or Canada and are curious, the free tier lets you test Muse without a commitment. A cautious way to start is to connect read-only access to one system, such as your calendar or a reporting export, and see how the agent performs before granting it write access to anything that can send messages or move money.

Before connecting accounting or payment tools, check which permissions each connector asks for and whether you can restrict it to reading. Look at the audit trail after each task to confirm what was actually done. If you already pay for a workflow automation tool or an assistant built into your store or accounting platform, compare how each one handles approvals, logs and data retention before adding another agent with access to the same records.

Outside North America, there is nothing to sign up for yet. It is worth watching for Meta's pricing and for its data processing terms if the service expands to your country.