Tutorial
Muse First Task Guide: From Login to Verified Output
A step-by-step onboarding for Muse, Meta's personal AI agent: log in, set ground rules, complete one real task, verify the output, and revise it.

Muse First Task Guide: From Login to Verified Output
Muse is Meta’s personal AI agent. It can browse the web, generate files, and keep some tasks running after you close the app. This guide walks you through a single real task, in order. Follow the steps, not the feature list.
1. Log In and Scope Out the Environment
Start at the official Muse signup page. Desktop, mobile, and web are supported. As of the source material, Muse launched first in the US for iOS, Android, and web. Availability depends on the signup page you see — check the official page for current regions and tiers. There are free and paid plans. Complete one task before deciding which tier fits you.
After login, do not start typing random requests. Take two minutes to understand the interface: where the chat input is, where task messages live, and where the activity log sits. You will need all three shortly.
2. Set Ground Rules Before You Work
Right after login, establish collaboration rules in plain language. Here is a practical starting set:
- You are a Chinese user in the Asia/Shanghai timezone.
- Answers should lead with the conclusion, then supporting detail.
- Anything unverifiable must be labeled with a source to double-check.
- Any operation involving external sending, form submission, ordering, payment, or account changes must be shown to you for explicit approval first.
Then go to Settings. Review three things:
- Permissions — what Muse can read, edit, or send.
- Approvals — which actions require your confirmation.
- Existing connections — which accounts are linked and what they can access.
Chat agreements do not replace permission settings. Sensitive operations go through an approval flow. When an authorization card appears, read the exact action before approving. Do not approve blind.
3. Run the Guided Practice Task
Now you will complete one visible task end-to-end. Ask Muse to organize two official announcements into a “Muse starter reference sheet.”
Use this task formula:
Task + source scope + completion criteria + delivery format
Example prompt:
Organize the two official Muse announcements I will paste below into a single reference sheet. Use only these two sources. Keep citations for each claim. If something is not readable, mark it as “not stated.” Do not sign up for anything or request payment. Deliver an openable table file and attach the file entry in this chat.
Paste the two announcements. Wait for the file.
4. Verify the Result
Do not trust the output. Open the file and check:
- Does it render correctly?
- Are headers complete?
- Are source links clickable?
Then spot-check three claims against the originals. Ask: Is this claim actually in the source? Is anything exaggerated? Click each source link and confirm the content supports the conclusion.
If a claim is unsupported, note it. Do not fix it yet.
5. Make One Revision in the Same Thread
Stay in the same chat. Do not start a new conversation.
Request a revision with these specifics:
- Keep the original file intact.
- Save a new version as v2.
- Add a “publish date” column.
- Re-check all claims against the sources.
- Move anything unsupported into a “to verify” section.
- Provide the new file entry plus a change list.
When the new file arrives, open it and repeat the spot-check on the changed rows.
6. Add Capabilities After a Few Uses
After a few tasks, you can expand what Muse does for you.
Memory stores your information and preferences. It improves through normal conversation feedback. If you correct Muse, it should update Memory. Review changes via the avatar → Identity page.
Soul shapes expression style. Tell it how to write, and refine through feedback.
Email — connect your email only if the material you need lives there. Disconnect when done.
Goals — turn multi-day items into goals. Set a deadline, list missing information, and define next steps. Muse can keep these running after you close the app.
7. Advanced: Custom Connectors and Computer Access
Custom connectors — the Settings list may include phone, browser, Gmail, finance via Plaid, Apple calendar/contacts/reminders, and Calendly. If something is missing, ask Muse for a configuration plan.
Never paste API keys, passwords, or verification codes into a plain chat. If a setup requires a secret, use the proper approval flow.
Connecting your computer — on Mac, get the client from the official entry and grant operation permissions. Before granting anything, ask Muse to explain the access method and permissions. Then test in a scratch folder. List filenames only. Do not read unrelated directories. Do not modify or delete anything.
8. Troubleshooting
- It only explains without delivering — restate the deliverable requirement. Say: “I need a file, not an explanation.”
- A file is lost — find the entry in task messages.
- Progress is unknown — check the activity log or ask “where is it stuck?”
- A preference is not applied — point it out, correct it, then check Memory.
- An operation looks wrong — deny the approval and disconnect the service.
Final Checklist
Before you close this guide, confirm all four:
- Logged in and know where to check permissions and activity.
- Issued a task with clear source scope and deliverable.
- Opened the deliverable and spot-checked sources.
- Requested a revision and confirmed the new version.
That is the full loop. Do it once, and you will know how to work with Muse.
(Figures and case details in this article are as reported in the source material and have not been independently verified.)
Series: AI Practice