How to Give Meta Muse Its Own Email Address

How do I give my Meta Muse agent an email address?

Message Muse: "Get yourself an AgentMail email address." Muse replies with a plan. You create an AgentMail API key named "Muse" in the AgentMail console, paste it into the Secrets tab Muse opens for you, and approve Muse's request to connect to AgentMail. Muse then creates its own inbox, a real address on agentmail.to that it can send from and receive at. Ask it to email your personal address to confirm it works. The whole thing takes about five minutes and needs no code, because Muse writes its own connector to AgentMail's API. AgentMail's free tier covers everything in this guide.

On this page

• How do I set up AgentMail on Muse?
• Where does the API key go?
• What is the approval section for?
• Can I choose the email address?
• What can Muse do with its own inbox?
• Gmail is not built for agents

How do I set up AgentMail on Muse?

You need a Muse account (US only at launch) and an AgentMail account.

1. Ask Muse for an address. In your main chat, send: "Get yourself an AgentMail email address." Muse replies with its plan: it will ask you to sign up for AgentMail, create an API key, and hand it over.
2. Create an API key. Go to the AgentMail console and sign up. Create a new API key, name it "Muse" so you can find and revoke it later, and copy it.
3. Paste the key into the Secrets tab. Muse opens a Secrets tab for the key. Paste it there. Don't paste API keys into the chat itself.
4. Approve the connection. Muse's request to share information with AgentMail appears in the approval section on the right of the app. Approve it there. You can grant a standing permission for the site so Muse doesn't ask on every call.
5. Let Muse create its inbox. Muse creates an inbox and picks the address and display name unless you told it otherwise. It tells you the address when it's done.
6. Send a test. Ask: "Send an email from your AgentMail address to me at you@example.com." Check your personal inbox. When it lands, the setup works.
7. Reply to it. Reply from your personal inbox, then ask Muse: "Did I reply?" It reads the reply in the same thread.

Where does the API key go?

In the Secrets tab, which stores it in Muse's secure credential store instead of the conversation. Meta's security design keeps credentials out of the agent's reach: the agent works with a stand-in token, and a separate service swaps in the real credential at the network boundary. A prompt injection that talks Muse into printing its secrets has nothing real to print.

Naming the key "Muse" matters later. If you ever want to cut Muse off from its inbox, delete that one key in the AgentMail console. Your other keys and inboxes keep working.

What is the approval section for?

Every call Muse makes to an outside service goes through Sentinel, a separate permission agent on Muse's VM that Muse can't override. The first time Muse talks to AgentMail, Sentinel asks whether Muse may share information with it, in the approval section on the right of the app. The request renders in the app UI, outside the chat, so a manipulated agent can't approve itself.

You choose how long the permission lasts. A one-time approval means Muse asks again next time. A standing permission for the AgentMail site lets Muse send, check its inbox, and create new inboxes without interrupting you. If you'd rather approve each send, keep the shorter grant.

Can I choose the email address?

Yes. If you don't specify one, Muse picks the address and the display name. To choose, ask: "Create an inbox called brooke" or "Create an inbox called car-sale with the display name Brooke." If the name is available on agentmail.to, Muse creates it. If it's taken, Muse tells you and you pick another.

Muse can hold more than one inbox. Muse organizes work into a main chat and side chats, and a separate address per side chat keeps each job's email in its own thread history: one for selling the car, one for the move, one for wedding vendors.

With a custom domain on a paid AgentMail plan, the address can be on your own domain instead of agentmail.to.

What can Muse do with its own inbox?

1. Contact businesses as itself. Quote requests to movers, a special request to a restaurant, a follow-up on a refund. Replies come back to Muse's inbox, not yours, and your personal address stays off vendor mailing lists.
2. Sign up for accounts. Muse's Gmail connector filters verification codes out of your inbox by design, so it can't finish signups there. With its own address, Muse signs up and reads its own code.
3. Work from what you forward. Don't want to connect your Gmail at all? Forward Muse the order confirmation, the school form, or the contractor thread, and it works from that alone.
4. Join a thread. CC Muse's address on an email thread. It reads the history and follows along.
5. Catch alerts. Price drops, restock notices, and promo emails go to Muse's address. It surfaces the ones that matter.

For copy-paste prompts for each of these, see /build/muse.

Gmail is not built for agents

Gmail was built for a person reading their own mail, not for automations and bots acting on their own. AgentMail is built for agents: inboxes created by software, owned by the agent, with full thread history. The two do different jobs, so keep the Gmail or Outlook connector if you want Muse to search and triage your own mail. It does that well, across multiple accounts. AgentMail covers the jobs the connector isn't built for: an address Muse owns, sends that don't come from you, and mail other people can send straight to Muse.

One thing not to do: don't ask Muse to open a browser and create a Gmail account for itself. Google's account policy prohibits using bots to create accounts, and agent-created Gmail accounts have been disabled within days. Why Muse needs its own address, and why not Gmail covers the details.