An AI that remembers your goals, and shows its receipts

Why “it remembers you” is usually a warning sign

Most memory features work like this: the assistant quietly forms an opinion about you, and from then on that opinion shapes every answer you get. You cannot see it, you cannot check it, and when it is subtly wrong the assistant becomes subtly and permanently wrong with it.

The failure is not that it remembered. It is that you cannot audit what it remembered.

What the companion actually writes down

StickyFriend’s companion keeps durable things it learns while you work: facts, your goals, your preferences, and reminders. Not the whole conversation — the parts that would still be true next month.

The part that matters: every item carries citations back to the exact turn it came from. An item is not “you prefer concise answers”; it is “you prefer concise answers”, plus a link to the moment you said so. You can open it and check. That is the difference between a memory you trust and a memory you verify.

Say goals out loud, in the shape you want them kept

The companion writes down what it learns, so how you phrase a goal decides how usefully it is stored. Some patterns that work:

• State the goal, then the constraint. “I am migrating this service by the end of the quarter, and I cannot take downtime” stores as one goal with the thing that actually governs every decision attached to it.
• Say when a goal is finished. Goals go stale silently. Telling it a goal is done is faster than wondering why it keeps steering toward last month’s project.
• Correct it in the moment. A correction is itself a turn, so the corrected item is cited to the correction. The trail stays honest.

Recall, and the boundary around it

Ask about something you discussed in a different conversation and StickyFriend can go and find it. That is genuinely the thing you want: the answer from three weeks ago, without you remembering which chat it was in.

It has a deliberate fence around it. Recall stays inside the current project unless you turn on cross-project recall yourself, and it ships off. The reason is not technical shyness. A project is how you separate the client work from the job hunt from the personal admin, and a memory system that quietly crosses that line has broken the only boundary you drew.

The controls, all four of them

• Switch memory off entirely. Not “pause”; off.
• Delete individual items. If one is wrong, it goes, rather than being outvoted by later evidence.
• Pin the ones worth keeping, so the things you actually rely on are not competing for space with trivia.
• Leave cross-project recall off. It already is.

One honest caveat, because it surprises people: long chats still get summarised even with memory switched off. That is not a memory feature sneaking back in — it is what keeps a very long conversation coherent and readable inside a single chat. It stays in the chat; it does not become something the companion knows about you.

Where the memory lives, and who sees it

Memory items are stored on your account, alongside your chats, projects and notes. Your chats are answered by our own self-hosted model by default; only if you switch a chat to the Smarter AI does a third-party provider see that chat. The screenshots the agent takes while working a page are not saved to the cloud unless you switch that on yourself, and that is enforced on the server rather than merely hidden in the interface.

The full accounting — what is stored, who touches it, how to delete it — is on the privacy page, including a plainly-stated weakness in one network hop that we would rather tell you about than let you discover.

The short version

Remembering is easy. Being checkable is the hard part, and it is the part that decides whether you can let an assistant carry your goals around. Ask any StickyFriend memory item where it came from, and it will tell you the turn.