You have an assistant. They have a mailbox. That is a normal pair, and it is the pair most first sends actually are: a lawyer, a contractor, a parent, a founder who has not connected anything. If the product required them to install a model before they could read you, you would not write them. You would wait, and the decision would wait with you.
RelayLink is built for that person. The assistant is optional on their side. The message is not.
They get mail, not a connector
A send to someone who is not on RelayLink travels as email — a briefing and a link. They do not create an account to open it. They do not paste a key. They do not join a workspace. They open a page, read labelled prose, and type a reply if they want to.
If they have never heard of the product, the notification says what it is and that receiving is free. It does not tell them to ask an assistant they have not connected. That line is for people who already have an inbox here.
Letting your AI email someone is this path with an approval step in front of it. You still confirm. They still get ordinary mail.
The page has to work without their model
They may print it. They may forward the link to a partner. They may read it on a phone between other mail. There is no script on that page, on purpose: it is unauthenticated, prefetched by scanners, and it renders a stranger's words. What they see has to be the whole message — ask, options, assumptions, note — without a chatbot to "summarise it for them."
Write like that. If the ask only makes sense after a session you had with your assistant, they will not have that session. They will have the page. The reply is the point. Design the page so a person with no model can give you one.
Attachments are out of scope. If the decision depends on a spreadsheet they must download, you are not done composing. Put the figure in the brief or wait until you can.
Their "no AI" is not a lesser path
It is easy to treat the magic-link reply as a fallback and the assistant pull as the real product. For them, the link is the product. Acknowledging is a button they press, not a tool call. A scanner fetching the page is not a read. You do not get to interpret a prefetch as engagement.
If they later sign in, they become a real account. After that, you cannot keep cold-sending. You ask them to be a contact. The gate is the same one that protects everyone else's assistant inbox. Their lack of a model never made them less entitled to that protection.
What you owe them
A finished ask. Labels so they can see which words are yours. An honest urgency. A note that does not assume they will "run this through Claude." If you need a decision by Thursday, say Thursday. If you need a call, say why writing will not do.
They can answer without joining your stack. That is the whole courtesy. Do not spend it on a message that only another assistant could parse.