You do not need Claude or ChatGPT to use RelayLink. Sign in at your account page with an emailed code, and you get the same inbox, contacts and send path your assistant would see — in a browser.
How you sign in
There is no password. Type your address, RelayLink emails a short code, you enter it, and you have a session.
That is also how a first-time recipient becomes an account holder. If someone only ever received a briefing by email, completing the code is what activates them.
The browser session is not the same as connecting an assistant. Claude and ChatGPT sign you in through OAuth when you add the connector. ChatGPT requires developer mode before its settings expose the custom-connector field. An agent framework uses an API key in a header. The portal uses the code.
What you can do there
Inbox. What is waiting for you — who sent it, the topic, whether it is a reply. Opening a package shows the full briefing and a reply box.
Threads. Conversations you are already in, including ones you started and ones that arrived as email.
Contacts. People you correspond with on a standing basis, requests waiting on you, requests you sent, and anyone you blocked. Nicknames live here too.
Compose. Start a briefing from the keyboard. You fill in who, the topic, your note, the ask, what kind of reply you want, and how urgent it is. An assistant would also structure assumptions and options; the form does not ask for those, because a person sitting down to write already knows what they mean.
Keys. Issue and revoke named credentials for clients that cannot sign you in. The secret is shown once.
Approvals. When your assistant has drafted something, /account/drafts/{id} is where you review the note on a session the assistant does not hold. Review a draft in the portal is why that page exists.
Help and settings. Support, your display name, connected applications, outbound webhooks where this host offers them, export and close.
Opening a page does not tell the sender you read it
When your assistant calls get_package, RelayLink records that their assistant fetched the briefing. That is what thread_status uses to answer "have they seen it."
Opening the same package in the browser is a different act. Mail scanners already prefetch the link in the notification email. If a fetch counted as "read," the sender's trail would lie. So a portal view is not shown to them, and neither is a magic-link fetch.
If you want them to know a person read it, press Acknowledge — or tell your assistant to. Acknowledging is not opening is the full split.
Compose is not a softer send
The portal does not get its own consent rules. Writing to another RelayLink account still needs an accepted contact. Writing to someone without an account still travels as email. A block still stops everything.
There is no approve-the-draft step on a typed package, because you wrote it. The label on those words is "typed by the sender," not an AI draft you rubber-stamped.
A reply you type in the browser lands on the same thread as one that arrived through your assistant or through email. It is one account.
The pages work with the script blocked
Forms post, links navigate, a success notice shows up after a redirect, and a refusal comes back on the same page with the reason above the form. The script adds a busy button, a confirmation sheet, local clocks. It is not required.
If you have not connected an assistant yet, sign in and send one briefing from compose. When you want the connector path, add RelayLink to Claude or ChatGPT.