Muse for Small Business: Put Meta's AI Agent to Work Without Losing Control

Most small business owners use AI as a chat box that answers questions. You type a prompt, get an answer, copy it into another tool, and move on. Muse for Small Business changes the unit of work. Instead of a single answer, you give Muse a goal. It keeps working on that goal in the background, across your connected tools, on its own secure computer, and brings back drafts, reports, and decisions for your approval. You get the value when you treat it as an agent that works for you, not a search engine that talks back.

This is the operator playbook for putting Muse to work without losing control of your accounts, your money, or your brand. You will learn how to start with one business goal and a written brief, connect only the systems that goal needs, use permission settings and approval cards as your real control layer, demand reviewable work, and run a weekly governance review. No fluff, no invented case studies, no promises of time saved. Just the implementation steps grounded in what Meta built and what works in practice.

What Muse Actually Is

How It Works

Muse is Meta's personal AI agent. It does not just answer questions. It actually does the work. It helps people stay on top of things, takes tasks and projects off their plate, and turns long-term goals into action plans. Muse runs on a dedicated secure computer with its own browser, file system, and terminal. Meta calls this the Muse Secure VM. The agent and your data live there, contained so no one else's agent can reach it.

You talk to Muse like you would message a person. You can use the Muse app, muse.ai, or message it directly in WhatsApp. Once you set a goal or task, Muse continues working on a schedule and in response to relevant events. It keeps working after you close the app. It comes back when something changes or when it needs your approval, like before it sends an email or makes a purchase.

Muse has one long main conversation and side chats for separate context. It remembers what matters. You can tell it to forget specific things. The Goals tab tracks your goals and plans. The Ideas tab offers suggestions. It produces documents, PDFs, web pages, trackers, and dashboards, which Meta calls Artifacts. You can tell Muse to turn proactive messages off, dial them down, or dial them up.

Muse for Small Business

On September 29, 2026, Meta expanded Muse with Muse for Small Business, a collection of new skills and connectors to help people run their businesses. Muse now connects to Instagram professional account analytics, Facebook Pages, and Meta ad accounts in a few clicks. Meta says it already understands your business: what you sell, what your brand sounds like, and what customers keep asking you about.

The named connectors include Asana, Box, Canva, Dropbox, Figma, Granola, HighLevel, Intuit QuickBooks, Klaviyo, Lovable, Notion, Shopify, Slack, Stripe, and Zoom, plus Facebook and Instagram business accounts. Available connectors may vary by account. Check the current list in Settings. You can also set up custom connectors for services not yet supported, though Meta does not review custom connectors, so treat them as higher risk.

Availability and Pricing

Muse is available to people 18 and older in the US and Canada, per Meta's business help pages. Check current availability before you plan around it. Muse is free for most of what people need, with Power and Maximum subscription plans for higher weekly usage limits. Confirm current plans in Settings or Meta's help pages.

Start With One Goal and a Real Brief

The difference between getting value and creating a mess is starting with one goal and a written brief. Do not connect everything on day one and ask Muse to figure out what you need. Pick one business goal, write down what you want, and give Muse the constraints and sources it needs to do the work without guessing.

The Brief Template

A goal brief has eight parts:

1. The goal itself: what you want Muse to accomplish.
2. The result you want: a draft, a report, a list, a summary.
3. The source material: which accounts, files, date ranges, or conversations Muse should use.
4. The audience: who will see or use this work.
5. The deadline: when you need it or how often.
6. The rules: what Muse should always do or never do.
7. The cadence: one time, daily, weekly, monthly.
8. The approval boundary: what Muse should not send, publish, spend, or share without asking you first.

Meta's business help pages lay out similar steps:

1. Tell Muse what you want and how often.
2. Describe the result, add source material, audience, deadline, and rules.
3. Set up recurring work like a weekly summary or daily check of messages.
4. Review Muse's plan and add missing details.
5. Review account access when a task needs another service.
6. Review approvals.
7. Check what Muse is working on.
8. Check finished work before using or sharing.

Picking a First Goal

Meta suggests starting with one of these prompt themes:

• Draft a growth plan from this year's sales, campaigns, and social performance.
• Triage your email, calendar, and news when you are underwater.
• Analyze your ads and content performance and draft next week's campaign for review.
• Run a monthly financial review that flags expenses that look off.
• Draft outreach to new customers and check in on ones who have gone quiet.

The best first goal is read-and-prepare work that ends in a draft or report, not something that publishes, sends, or spends. You want to see how Muse interprets your brief, what sources it uses, and how it handles facts and numbers before you let it touch anything customer-facing or financial.

Example Brief

Here is one example brief for a small service business:

• Goal: weekly pipeline review and follow-up prep.
• Result: a summary of open opportunities, which ones moved, which ones stalled, and draft follow-up messages for the stalled ones.
• Source material: opportunities in your CRM with activity in the last 30 days, email threads tagged with the client name, calendar events from the last two weeks.
• Audience: you and your sales lead.
• Deadline: every Monday by 8 a.m.
• Rules: only include opportunities with a value over a certain threshold, do not include internal test accounts, use the client's name and the last interaction date in every draft message.
• Cadence: weekly.
• Approval boundary: do not send any message, do not update any CRM field, do not create any calendar event. Bring everything back as a draft for review.

Write your brief in a document before you give it to Muse. Read it out loud. If another person on your team could not execute it, Muse will struggle too. Once the brief is clear, paste it into Muse and ask it to confirm what it understood. Tighten any part that Muse misread. Ask Muse to remember the final brief so it can refer back to it.

Connect Only What the Goal Needs

Connector scoping is the practice of connecting only the systems your goal requires and nothing more. Every connector you add is another place Muse can read data, take actions, or make mistakes. Start narrow. You can always add connectors later when a new goal needs them.

Map your goal to the minimum set of connectors. If your goal is a weekly pipeline review from your CRM and email, connect your CRM and email. Do not connect your accounting system, your ad account, or your file storage yet. If a later goal needs QuickBooks, connect it then.

Read-Only First

Many connectors can be set to read-only, which means Muse can retrieve data but not take actions. Use read-only mode while you are learning how Muse interprets your goals and what it tries to do. You can switch to read-and-write access later when you trust the pattern. Check the connector settings in the Muse app to see which connectors support read-only mode.

Meta Connectors

Facebook, Instagram, and Threads connect automatically if they are in the same Accounts Center. You do not need to connect them separately. Muse for Small Business can use Instagram professional account analytics, Facebook Pages, and Meta ad accounts alongside the other services you connect. Make sure you are signed in to the Meta account that owns or has access to the business assets you want Muse to work with.

Custom Connectors

Custom connectors let you connect services that are not yet in the official list. Muse guides you through the setup, which may involve retrieving API information from the service. Meta does not review custom connectors. Treat them as higher risk than the official connectors. Only connect custom connectors when you trust the service and you understand what data the connector will exchange.

Disconnecting and Forgetting

Disconnecting a connector stops future data exchange, but information Muse previously used may remain in memories and conversation history. If you disconnect a connector because you no longer want Muse to have that data, also ask Muse to forget the information it already pulled from that connector. You can do this in the conversation, and you can read and edit Muse's Memory files directly.

Why the Architecture Matters to an Owner

Understanding how Muse is built helps you know where the real control points are and why approvals and least privilege still matter even though Meta built a lot of safety into the system.

The Secure VM

Muse runs on a dedicated cloud computer, the Muse Secure VM. This is an isolated Linux box with a browser, file system, and terminal. Your data and the credentials for your connected services are stored there. The VM is the system of record. You can inspect, edit, and download the files and memory. Meta runs continuous backups.

Sentinel and Scoped Approvals

Sentinel is a separate agent that acts as the sole permission authority for connector actions and network egress. Muse proposes an action. Only Sentinel grants it. Sentinel decides to allow, deny, or ask you based on your connector policy and permission settings. Approval dialogs come in the client UI, not in the conversation. This separation means Muse cannot talk itself into permissions it does not have.

Approvals are strict capabilities bound to the connector, the destination, and the use case. They can be one-time, session-scoped, task-scoped, time-bounded, or perpetual. The point is not to ask you about everything. Read-only actions, previously allowed actions, or demonstrably low-risk actions can proceed without interruption. But anything that publishes, sends, spends, or shares stops for approval.

Credentials and Logins

Muse never sees your real API keys or passwords. Credentials for the services you connect are kept in a separate secure store inside your VM. Code that Muse runs only ever sees a stand-in surrogate token, and the real credential is swapped in at the network boundary after the request is authorized. If someone tries to trick Muse into revealing your secrets, it does not have them to reveal.

The email connector filters out one-time passcodes, password reset links, and login magic links. Muse will not see or use those messages. Meta's reasoning: your inbox can reset passwords across most of your other accounts, so connecting email should not let the agent, or someone manipulating it, represent you across other sites.

You can watch what Muse is doing in its browser and take over at any time. When you take over, the agent is paused. When you sign in to a website for Muse, the app captures your username and password in a separate login screen and sends them straight to secure storage that the main agent cannot see. The credential is filled into the browser only at the point of need.

Purchases

For purchases, Muse asks for human approval with exact purchase details every time. Meta uses Link by Stripe to generate a one-time card tied to the merchant, the amount, and a limited time window. Link purchase protections apply to eligible purchases. Shop Pay support is coming soon but is not available yet.

Prompt Injection

Meta has documented how Meta built safety into Muse in detail. The architecture uses defense in depth: model training, labeling untrusted input, classifiers, and human approvals for actions that move data out of the VM. But Meta also says prompt injection remains an open problem in the industry, and Muse will sometimes make mistakes. This is why approvals and least privilege still matter. The architecture reduces risk. Your permission settings and approval decisions are the last line of defense.

Permissions and Approval Cards Are Your Control Layer

Permission settings and approval cards are where you actually control what Muse can do. Treat them as the real control layer, not an annoyance to click through.

Permission Settings

Muse distinguishes between read actions and write actions. Read actions retrieve data. Write actions change data, send messages, publish content, or spend money. By default, Muse will not take many important actions, like sending an email, without approval. You control how cautious Muse is in Settings under Permissions.

For Connectors, you have two default modes:

• Ask for some actions means Muse asks before every write action and important read actions.
• Always ask means Muse asks before any action, read or write.

Start with Always ask while you are learning what Muse tries to do. Once you understand the pattern, move to Ask for some actions. You can also set permissions per connector, so you can be more cautious with your accounting system and less cautious with your file storage.

For Web access, the defaults work the same way:

• Ask for some actions means Muse asks when your information may be shared or when it visits an unfamiliar site.
• Always ask means Muse asks before any website.

Use Always ask at first. As you see which sites Muse uses regularly and trust the pattern, you can move to Ask for some actions.

Approval Options

When Muse asks for approval, you see a structured approval card that describes the action. You have five options:

• Allow once means Muse can take this action right now, one time.
• Allow for this task means Muse can take this type of action for the entire task.
• Allow for this site means Muse can take this type of action on this website in the future without asking again.
• Always allow means Muse can take this type of action for this connector in the future without asking again.
• Deny stops the action.

Tap See task details before you decide. The details give you more context about the task before you decide what permission to grant. Do not approve a card you cannot explain. If you are not sure, deny it and ask Muse to explain what it was trying to do and why it needs that permission. Tighten your brief or your rules if Muse is asking for permissions that do not match your goal.

The Permission Ladder

Use a permission ladder as you learn what Muse does and build trust. Start with Always ask for everything. This is the most cautious setting. You will see every action Muse wants to take. It will feel like a lot of approvals at first. That is the point. You are learning the pattern.

After a few days, move to Ask for some actions for connectors and web access. Now Muse can take low-risk read actions without asking, but it still stops for writes and important reads. Watch the Activity log to see what Muse is doing without approval. If you see actions you do not like, tighten the rules in your brief or switch back to Always ask for that connector.

When Muse asks for approval, use Allow once and Allow for this task before you use Allow for this site or Always allow. Allow once and Allow for this task are narrow grants. They expire when the action is done or the task is finished. Allow for this site and Always allow are broad grants. They stay in place until you revoke them. Keep broad grants for low-risk, read-type actions. Keep publishing, sending, spending, and sharing behind per-action approval.

Avoid banner blindness. Banner blindness is when you approve everything just to make the dialogs go away. If you find yourself approving cards without reading them, you have moved too fast. Step back. Switch to Always ask for a week. Rebuild your understanding of what Muse is doing and why. Then move forward on the ladder again, slower this time.

Irreversible Actions

Meta says nothing publishes, sends, or spends without your approval. That is true. But Meta's help pages also say some actions, like sending an email, cannot be reversed. Approval comes before the action. Once you approve and Muse takes the action, you may not be able to undo it.

This is why the evidence standard matters. Check the work before you approve the action. Read the email draft before you approve sending it. Check the account and the amount before you approve a purchase. Review the post before you approve publishing it. Once it is sent, published, or spent, you own the outcome.

Demand Reviewable Work

Muse produces Artifacts: documents, PDFs, web pages, trackers, dashboards. Every goal should end in a reviewable artifact, especially while you are learning how Muse interprets your briefs and handles your data.

The Evidence Standard

When Muse finishes a task, open the Artifact and check it against an evidence standard:

• What sources did Muse use? Ask Muse to list the sources if they are not already in the Artifact.
• Are the facts accurate or are they assumptions? Check facts, prices, names, and numbers against the original source. Do not trust Muse to get numbers right without verification. If Muse pulled data from a connected service, open that service and confirm the numbers match.
• If Muse drafted content, check the voice and the claims. Does it sound like your brand? Are the claims true? Are the links correct?
• If Muse drafted an email or a message, check the recipient, the subject, and the tone. Is this going to the right person? Is the tone appropriate? Would you send this yourself?
• If Muse made a change in a connected service, check which account the change landed in. If you manage more than one account, this matters. Make sure the draft, the update, or the post went into the right account. If it went into the wrong account, undo it if you can, or fix it manually, and tighten your brief to specify the account next time.

Check what is waiting for approval. Muse may have finished part of a task and paused because it needs your approval to continue. Check Upcoming, which you reach by tapping your Muse avatar, to see scheduled tasks and reminders. Open the Activity log to see what Muse has already done and what permissions you have granted. If something looks wrong, stop the task, revoke the permission, and ask Muse what happened.

When Muse Stops

Sometimes Muse stops in the middle of a task. Meta's troubleshooting steps are straightforward:

1. Ask Muse what it completed, what stopped it, and what it needs.
2. Check the status of the task.
3. Check connection access in Settings under Connectors. If Muse lost access to a connector, reconnect it.
4. If Muse is missing information, add the missing source, format, deadline, or rule to your brief.
5. Then ask Muse to continue.

If Muse stops repeatedly on the same task, the brief is probably too vague or the goal is too broad. Break the goal into smaller tasks. Give Muse one task at a time with a clear result and a short deadline. Once Muse completes the smaller tasks reliably, you can combine them into a larger recurring goal.

Recurring Work Without Surprises

Recurring work is where a clear brief pays off. A weekly summary, a daily check of messages, a monthly financial review. You set it up once, Muse runs it on schedule, and you review the output.

When you set up recurring work, be explicit about the cadence and the approval boundary. Tell Muse how often to run the task: daily, weekly, monthly, every Monday morning, the first of every month. Tell Muse what it should not do without asking: do not send, do not publish, do not spend, do not update any record. Bring everything back as a draft or a report for review.

In Settings, Permissions, Manage permissions, the Artifacts and scheduled tasks option lets you manage approvals for each scheduled task and for artifacts. If a recurring task always needs the same write action, like saving a draft in a connected tool, you can grant Allow for this task so Muse does not ask every time. But keep publishing, sending, and spending behind per-action approval even for recurring tasks.

Upcoming (tap your Muse avatar) shows scheduled tasks and reminders. You can cancel scheduled work whenever you want by telling Muse. If you change a recurring task, update the brief in the conversation.

Proactive messages are how Muse notifies you when something is meaningfully new or needs your input. You can tell Muse to turn proactive messages off, dial them down, or dial them up. If Muse is interrupting you too often, dial them down. If you want Muse to notify you about more events, dial them up. You stay in control of when and how Muse gets your attention.

Where Muse Fits in Your Stack

Muse is an agent that works for the owner. It is not a customer-facing chatbot. Understanding where Muse fits in your stack helps you avoid overlap and use the right tool for each job.

Front of House vs Back Office

Customer-facing automation, like answering leads, booking appointments, and CRM follow-up, is a different job. Meta Business Agent is the Meta tool for messaging customers, capturing leads, booking appointments, and supporting sales, on WhatsApp and Messenger, with Instagram Direct listed as coming soon. If you need an AI agent inside your CRM to qualify leads, update records, or trigger campaigns, that is a job for AI agents built into or integrated with your CRM, such as GoHighLevel Managed Agents.

Muse works for you, behind the scenes, on internal tasks. It drafts content, analyzes performance, summarizes data, prepares reports, and brings decisions back for your review. It connects to the tools you use to run the business: your ad accounts, your analytics, your file storage, your accounting system, your project management tool. It is not built to be the always-on front line for customer conversations. When it drafts outreach, you approve what goes out.

Meta AI Business Assistant is a separate tool for analyzing Meta ad campaigns, resolving common account issues, and getting recommendations and benchmarks, available across Meta advertising surfaces such as Ads Manager. Muse can also look at ad performance, but it works across all your connected tools and keeps working on goals over time.

Think of Muse as your internal operator. Customer-facing agents handle the front of the house. Muse handles the back office. When you keep that boundary clear, you avoid asking Muse to do work it was not designed for, and you avoid gaps where neither tool is covering a job that needs to be done.

Weekly Governance Review and Offboarding

A short weekly routine keeps Muse working the way you intended and catches problems before they grow.

1. Open the Activity log. Scroll through the actions Muse took in the last week. Look for actions you did not expect. Look for patterns that do not match your goals. If you see something you do not like, revoke the permission in Settings, ask Muse what happened, and tighten the rules in your brief.
2. Check Upcoming. Check the scheduled tasks and reminders. Cancel anything you no longer need. Update the brief for anything that has drifted from your original intent.
3. Open Settings and review Connectors. Disconnect any connector you are no longer using. If you disconnected a connector, ask Muse to forget the data it pulled from that connector.
4. Review Allowed websites. Revoke access to any site Muse no longer needs. If you see a site you do not recognize, ask Muse why it needed access and decide whether to keep it or revoke it.
5. Review your permission defaults. Are you still on Always ask, or have you moved to Ask for some actions? Are your per-connector permissions still appropriate, or do you need to tighten or loosen them?
6. Check Memory files. Muse stores information it remembers in Memory files you can read and edit. If Muse is remembering something you no longer want it to remember, edit or delete that Memory file.
7. Review your training decision. Meta says conversations and the resulting tool calls are sanitized to remove key personally identifiable information before they are used to train its models, and you can opt out with a switch in Muse settings. If you handle sensitive business information, decide deliberately.

Offboarding

If you stop using Muse, offboard properly:

1. Disconnect all connectors in Settings under Connectors. This stops Muse from accessing your accounts.
2. Then use Reset to permanently delete all your information, chat history, files, and active tasks. Reset is permanent. There is no undo. Make sure you have downloaded any Artifacts or Memory files you want to keep before you reset.

Common Mistakes

Here are the mistakes operators make most often when they start using Muse for business work:

• Connecting everything on day one. You do not need to connect every tool you use. Connect only what your first goal needs. Add connectors as new goals require them.
• Using Always allow on write actions. Always allow stays in place until you change it. Use it for low-risk read actions. Keep write actions, especially publishing, sending, and spending, behind per-action approval.
• Vague goals with no deadline or rules. Muse needs constraints. If your goal is vague, Muse will guess. The guess will not match what you wanted. Write a specific brief with a clear result, a deadline, and rules.
• Using drafts without checking numbers. Muse can make mistakes. Check every number, price, and account detail against the source before you use the work.
• Letting Muse post or email from the wrong account. If you have multiple accounts connected, specify which account Muse should use in your brief. Check which account the work landed in before you approve publishing or sending.
• Forgetting that actions Muse takes on the web count as your activity. When Muse browses or acts on a website for you, it appears as your activity, and Meta notes this can indirectly influence the ads you see. If you do not want Muse browsing certain sites, revoke web access or set stricter web permissions.
• Treating custom connectors like reviewed ones. Meta does not review custom connectors. They carry more risk than official connectors. Only use custom connectors when you trust the service and understand the data exchange.
• Never reviewing the Activity log. The Activity log is your audit trail. If you never look at it, you will not catch mistakes or drift until something breaks. Review it weekly.

30-Day Rollout Plan

Here is a recommended 30-day rollout plan. This is not a promise of results. It is a structured approach to learning how Muse works and building trust before you grant broader permissions.

Week 1: One Goal, Read-Only, Always Ask, Drafts Only

Pick one business goal that ends in a draft or a report. Write a full brief with result, sources, audience, deadline, rules, cadence, and approval boundary. Connect only the minimum connectors that goal needs. Set those connectors to read-only if the option is available. Set Permissions to Always ask for connectors and web access. Tell Muse the goal and the brief. Review Muse's plan. Let Muse run the task. Review every approval card. Check the Artifact against your evidence standard. Do not approve any action that publishes, sends, or spends. At the end of the week, review the Activity log and check what Muse did.

Week 2: Tighten the Brief and the Evidence Standard

Run the same goal again. This time, tighten the brief based on what you learned in week one. Add rules for anything Muse got wrong. Add sources for anything Muse guessed at. Raise your evidence standard. Check every fact, number, and account detail before you approve. Add one recurring weekly summary: a simple read-and-report task like a summary of your email, your calendar, or your social media performance. Set it to run every Monday morning. Make sure the approval boundary says do not send, do not publish, do not spend.

Week 3: Move to Ask for Some Actions, Allow Once and Allow for This Task

Move your permission defaults from Always ask to Ask for some actions for connectors and web access. Continue running your original goal and your weekly summary. This week, add one narrow write action. For example, let Muse save a draft in one connected tool you already use. When Muse asks for approval, use Allow once the first time. If the action works correctly, use Allow for this task the next time. Do not use Always allow yet. Review the Activity log mid-week and at the end of the week. Check that the write action landed in the right place and did what you expected.

Week 4: Weekly Governance Review, Decide What Earns Broader Permissions

Run your weekly governance review for the first time. Activity log, Upcoming, Connectors, Allowed websites, permission defaults, Memory files. Decide which actions have earned broader permissions and which actions stay behind approval. If Muse has been saving drafts reliably in the right place, you might grant Allow for this site or Always allow for that specific write action. If Muse has been reading your email reliably without problems, you might leave that on Ask for some actions. If Muse tried to do something you did not expect, revoke the permission and tighten the brief. The aim by the end of week four is one or two goals you can review with confidence, a weekly summary you trust, a permission ladder that matches your risk tolerance, and a governance routine you can maintain. Your timeline depends on your tools, your data, and how clearly you define the work.

Where SumoGrowth Helps

SumoGrowth is an AI implementation and consulting agency. We help small business owners, operators, and teams put AI agents like Muse for Small Business to work on real business tasks without losing control. We write goal briefs, scope connectors, set permission policies, and build review routines. No hype, no invented results, just implementation grounded in what works.

by Darius Gaynor