Give your AI assistant memory that survives the session

Your assistant forgets your architecture, your conventions, and yesterday's debugging every time a session ends.

When you are done: An assistant that can write something down and find it again next week.

Steps

Open the installer and pick your client.
Cursor and VS Code install in one click; everything else gets config to paste.
Decide whether the server runs locally or hosted.
Local needs Node.js 18 or newer. Hosted needs nothing installed.
Add a key only if you want a private namespace.
Without a key it still works, at a lower rate limit. With one, search and list cover only your own pastes.
Restart the client and ask it: "Save the text hello to Pastepile and give me the URL."
A URL back means the connection works.
Tell your assistant, in its own instructions, when to save.
Assistants rarely save unprompted. Naming the moments that matter is what turns this into memory.

Anything your assistant saves without a key is stored as plain text we can read. Do not have it save credentials or customer data that way.
An assistant cannot read an end-to-end encrypted paste, by design. The key is in a link fragment the server never sees, so encrypted content stays opaque to the model.
A key gives your assistant a private namespace, not a private server. Keep it out of anything you commit.