Muse Automation Recipes: 12 Workflows People Actually Run

The Muse automations that work best combine one connector, one clear stopping rule and, for anything recurring, a scheduled task. The twelve recipes below come from real builds in the catalog, each with what to connect, how the request is shaped, and the outcome the author reported. None of them needs code.

Numbers are as reported by the authors. Muse's quotas are counted in tokens and Meta has not published how many tokens a task uses (Layer3 Labs), so large jobs like archive scans may hit the free tier's limit.

Money recipes

1. Audit six months of receipts

• Connect: Gmail or Outlook.
• Ask for: every recurring charge in email receipts over the last six months, grouped by merchant, with anything you may have forgotten flagged.
• Reported result: Idobn's run found a forgotten $100/month subscription on a corporate card, $1,200 a year.

2. Search unclaimed property in every state

• Connect: nothing required; the state databases are public websites.
• Ask for: a search of each state's unclaimed-property database under your name and every city you have lived in.
• Reported result: $88.92 found. For a standing version, Dowser is a read-only connector that surfaces settlements, recalls with refunds, unused card credits and expiring points.

3. Negotiate a bill with a price ceiling

• Connect: whatever access the provider needs; the author authorized Muse to negotiate on his behalf.
• Ask for: a lower rate, with an explicit rule such as "accept any offer under $70/month."
• Reported result: Chandra Bhavanasi's AT&T fiber went from $80 to $40 a month plus 3 months free. The ceiling is the important part; it lets the agent say yes without checking back.

4. Pay a bill with a gift card

• Ask for: confirmation that a gift card can be applied to an account balance, then the payment.
• Reported result: Muse applied a Verizon gift card and paid down the balance, saving the user $100.

Inbox and social recipes

5. Screen unread Instagram DMs without opening them

• Connect: Instagram, which links automatically through Accounts Center.
• Ask for: a summary of each unread thread, and tell it not to mark anything read.
• Reported result: 50 unread threads screened, all still unread afterward.

6. Find dead follows

• Connect: Instagram.
• Ask for: the last activity date of every profile you follow, and a list of inactive ones.
• Reported result: a list of inactive profiles to unfollow. Muse surfaces them; you decide.

7. Count a huge archive with a resumable loop

• Connect: Instagram.
• Ask for: a paged count of your story archive that saves its cursor to a file every 10 pages, so a restart continues where it stopped.
• Reported result: a census of a 77k-story archive. The save-state instruction is reusable for any long job.

8. Build a content studio with a weekly trend check

• Connect: Instagram.
• Ask for: your last 100 posts clustered into content lanes, an ideas dashboard, and a weekly scheduled task that checks trends.
• Reported result: a working studio and cron. To publish as well as plan, hook up a scheduler like Postiz as a custom connector.

Daily briefing recipes

9. A new AI-news podcast every morning

• Ask for: a Muse podcast on the day's AI news, set as a scheduled task.
• Reported result: David Singleton gets a fresh episode each day for his commute. Peter Yang's 10 use cases video includes a similar personalized morning briefing.

10. Family schedule from school and sports apps

• Connect: the school and sports apps' feeds.
• Ask for: grades, announcements and game schedules in one place.
• Reported result: set up in about 30 minutes, versus 5 to 6 hours with another agent, according to the author. How I AI had Muse produce a one-shot family morning newsletter as a PDF.

11. A feed tuned to your portfolio

• Connect: Gmail.
• Ask for: your interests and holdings inferred from email, then used to fill your Muse feed.
• Reported result: Mostly Borrowed Ideas got a portfolio-aware feed. For account data rather than email, the Plaid connector covers 12,000+ US institutions.

Bulk file recipes

12. Merge hundreds of docs, or re-sort big playlists

• Connect: Google Docs, or Spotify.
• Ask for: one output file, or exact sorting rules.
• Reported result: 400 Google Docs became one 400-page PDF, plus HTML, Word and TXT versions. Another user re-sorted 18 Spotify playlists of about 200 songs each by decade and genre. The same user's attempt on 40k Google Photos failed, which is a useful ceiling to know about.

What the recipes have in common

Looking across all twelve, a few habits repeat:

• State the stopping rule. "Under $70/month." "Don't mark read." "Save cursor every 10 pages." The agent does better with a limit than with a goal alone.
• Start read-only. Connectors can be held to read-only in settings (Meta help). Recipes 1, 2, 5, 6 and 7 never need write access.
• Use scheduled tasks for anything daily or weekly. Recipes 8 and 9 only pay off because they repeat.
• Keep approvals on for money. Muse asks before purchases and sending email by default. Leave that alone until you trust a recipe.

For more ideas, the community-maintained awesome-meta-muse list catalogs 156 use cases with links to the original posts, and Agents Breakroom sorts about 500 by goal. Browse triage and routing builds for more of this kind, or read real errands people delegated.

Frequently asked questions

Can Muse run tasks on a schedule?

Yes. Builders pair tasks with scheduled runs, such as David Singleton's daily AI-news podcast and a weekly Instagram trend check. The agent keeps working in the background after you close the app.

What is the easiest Muse automation to start with?

A read-only audit is the lowest-risk start: have Muse go through six months of email receipts and list recurring charges. It needs only an email connector, changes nothing, and has produced real savings in reported runs.

Do Muse automations cost money?

The free tier is usage-limited, reported at up to 100M Muse tokens a week. Power is $20 a month for 500M and Max is $100 for 3B. Meta has not published how many tokens a typical task uses.

Can Muse handle very large jobs like thousands of photos?

Sometimes. A 77k-story Instagram census worked with a resumable, cursor-saving prompt, but one user's attempt to sort 40k Google Photos failed. Break large jobs into saved checkpoints.