RelayLink does not have a password. You type an address, a code arrives in that mailbox, you enter it, and you are in. That is sign-in, and on a host that is open to new accounts it is also how you get an account.
What the code is for
The code proves you can read that mailbox right now. It is not a standing credential. It is not an API key. It is not what Claude holds after you approve a connector.
Three different doors, one account:
- The browser — the emailed code, then a session cookie. Inbox, contacts, compose, keys.
- Claude or ChatGPT — the app sends you to RelayLink, you enter a code, you approve the connection. After that the app uses a token, not the code.
- A key — for agent frameworks and anything that configures a header. You issue that from your account page after you can sign in.
ChatGPT requires developer mode before its settings expose the custom-connector field. That gate is in ChatGPT, not in RelayLink's sign-in.
Signing in does not connect an assistant. Connecting an assistant does not replace the browser.
First visit, or a return
On a host that accepts new accounts, an address that is not yet in RelayLink still gets a code. Entering it activates the account. That is the same shape a cold send already used: a provisional row, then a real one once someone proves the mailbox.
On a host that does not, an unknown address gets no code and no hint. The page says an address without an account is on the founding list and will be written to. Accounts there still come from a cold send, a contact request, or whoever runs the deployment.
Which of those you are on is a fact about the host. The sign-in page says so. The setup guide cannot, because it is the same text everywhere.
Known and unknown look the same
Type a real account holder's address or a stranger's: same page, same timing. The verify step is padded too — an unknown address would otherwise come back faster, and that gap is enough to ask "does this person have RelayLink?"
An address that is not a plausible mailbox gets the same page and writes nothing. The browser's email field is a courtesy. The server does not create a row for junk.
What signing in activates
A provisional recipient — someone who only ever received a briefing — becomes a real account when they complete the code. After that, strangers cannot keep cold-emailing them into an assistant's inbox. Threads they already participate in stay open; a new thread from someone they have not accepted does not. A block still ends a conversation that was already going.
The welcome email goes out once, on that first activation. It points at /docs for the address and how this host authenticates, rather than restating either — the mail is static, the posture is not.
What this is not
It is not a magic link in the email. You copy a code; you do not click through to sign in. The package link is a different token, for reading one briefing, and it does not sign you into the account.
If you have never signed in, start at your account page. If you want an assistant connected afterwards, add RelayLink to Claude or ChatGPT.