EveryAI · the Muse field guide
Put your Muse
to work.
300+ things people actually ran with Meta's personal agent — 250+ with the exact prompt behind them. Hand the whole pack over in one paste, or steal a single prompt.
- Muse cases
- 0
- With the exact prompt
- 0
- Categories
- 0
- Reusable loops
- 0
01 The library
Find the job.
Copy the prompt.
Every case is real — pulled from the field guide, a crowd feed and cases people sent in, merged and de-duplicated every 6 hours. Community bot prompts from botdirectory.ai live in the Bots tab.
02 Start here
One prompt sets up the whole relationship.
Trying @muse, @bot or any other personal assistant? Connect the accounts you want it to reach — calendar, and email too if you yolo like me — then paste this. It covers onboarding, memory and the weekly review.
- First it briefs you Checks what's connected, reads 90 days of mail and calendar, and separates facts from guesses and gaps.
- Then it learns Keeps durable memory of your people, preferences and delegation rules — and self-reviews weekly so it stops repeating a miss.
- It stops where it should No messages, events or commitments on your behalf during onboarding. Afterwards it acts inside boundaries you agree together.
03 Field manual
Run one case,
or compose your own.
You don't have to hand over the whole pack. Every card carries the
exact prompt that ran the job — copy it, swap the
[bracketed] details for yours, send it.
- Filter to the job you want done — or search for it.
- Copy the prompt off the card.
- Swap anything in
[square brackets]. - Paste it in. Muse fills the gaps, then runs it.
The twelve loops
Most “new” use cases are combinations of these shapes. Compose them and you can invent cases that aren't in the library yet.
Guardrails worth writing in
Caveats that showed up again and again in hands-on reports. Put them in the prompt before you hand over anything important.
04 Dispatches
Ran something good?
Send it in.
Reviewed cases land in the library and in the capability pack on the next sync — with your handle on it if you want the credit.
- What counts Something you actually ran, not a hypothetical.
- The prompt is the gold Paste it verbatim if you have it. No prompt? Describe it and link the post.
- Reviewed before it ships Every submission is read by a human. Spam isn't.
Prefer git? The library and the worker are open source — send a pull request instead.
05 Open source
Fork the whole operation.
The page, the sync worker and the capability pack are MIT licensed. Run your own instance, point it at your own sources — or send a case in as a pull request.
everyai-com/muse-use-cases
Worker, page, build scripts and docs. One command to deploy your own.
github.com → LicenseMIT
Use it, fork it, ship it commercially. Attribution appreciated, not required.
Read the license → Public APIRead the whole library
/api/use-cases for JSON, /muse/capabilities.md for the
pack, POST /mcp for agents.
Send a case by PR
Add the case, run the build, open the pull request. Reviewed like any other.
How to contribute →git clone https://github.com/everyai-com/muse-use-cases cd muse-use-cases/worker && wrangler deploy node scripts/build-muse-site.mjs && wrangler pages deploy dist/muse
06 Provenance
Nothing here is invented.
Every case traces back to a post, a guide or a person — and the sync worker re-reads the sources every 6 hours.
Last sync: —
Prompts are community- and user-generated and unverified. They're starting points, not guarantees — read one before you run it, keep payments and account changes under your own approval, and be clear about what your agent may read, send or buy. Some newer cases arrive before they have a published prompt; those are marked and link to the original post.