Yeonghyeon ·

How do you choose where your AI's long-term memory lives?

Two things actually differ between the options: what gets counted, and what the assistant can reach. Here's how to check both.

The short answer

There are plenty of ways to give an AI a long-term memory, and most of them describe themselves the same way. Two things actually differ.

One: what gets counted. Some count how much you've stored. Others count how often you pull it back out. Some say "unlimited" and then sell you credits.

Two: what the assistant can reach. With some, connecting it means the assistant can see your whole workspace. With others, it only sees what you chose to keep.

Price and polish matter too, but those you can see at a glance. These two you can't.

Does this sound familiar?

Try a few and it tends to go like this:

What actually gets counted

This is the difference that's hardest to see.

A lot of them count retrieval. A free tier caps how many times a month you can look something up, or you buy credits, or it's metered by how much text gets processed. The names differ. What they have in common is that the meter moves when you use it, not when you fill it.

Which is why "unlimited" can mean two different things. Unlimited storage with a finite number of lookups is a common shape.

There's a reason it's built that way. Those services run an AI pass of their own when you save — pulling facts out of what you wrote, linking items to one another, building a graph out of them. Every one of those costs real money. When the cost lands on each operation, counting operations is the honest way to charge.

There's also a way to skip that work entirely: leave the tidying to the assistant you're already talking to. If it hands over a title and a summary already written, whatever receives it only has to hold on to it. Then the only ongoing cost is storage — so one number, how much you've stored, lines up with the cost, and there's nothing to be gained by counting lookups.

That's the side Contextick is on. One capacity number, and retrieval isn't counted.

What to check: find the unit on the pricing page before anything else. If the unit is how much you've stored, using it more doesn't cost more. If it's operations, lookups or credits, it does. Which of those is better for you comes down to how often you actually pull things back out.

What the assistant can reach

The second difference is what an assistant can see once it's connected.

Notion opens the whole workspace. Once its MCP is connected, Notion's own documentation says the tools "can access everything you can access." There's no step where you pick which pages to open, and search reads the contents of pages, not just their titles. That's convenient — it means everything already sitting in there is usable straight away.

The other shape opens only what you left there. Nothing exists except what you saved, so that's all the assistant can see either. The trade is that you have to leave it there first.

Which one is right depends on what's in that space. If you already work in Notion and what's in there is mostly work, connecting an assistant there is the right call — the material is already in place and nothing needs moving. If there's also things in there that aren't work — personal notes, things involving other people, drafts that aren't ready — narrowing the range is easier to live with.

What to check: look for a step in the connect flow where you choose how much to open. If there isn't one, all of it is open.

What doesn't actually differ

Some things get written up as differences and aren't.

"Works across several AIs" isn't ours alone. Now that MCP is a common standard, workspace tools and AI memory services alike connect to most assistants. That isn't a point of difference for us.

(Against an assistant's built-in memory it's a different story — that one works only inside that assistant, which is its own post. And what gets kept automatically isn't the same as what you chose to keep: the automatic side keeps getting broader, but you don't decide all of what stays.)

"How many items can I store" mostly doesn't differ either. Counting items is rare now. That's how old note apps worked, and comparing on it means looking at the wrong thing.

So which one fits you

The Notion side got answered above. What's left is how often you pull things back out.

If you retrieve rarely and mostly want to put a lot in, a service that counts lookups is fine. The free tier will feel generous.

If you want an assistant referring to it several times a day, check the unit. That's the usage pattern where per-lookup pricing shows up fastest.

If you'd rather not spend time filing things, and you want what the assistant can see limited to what you left there, something shaped like Contextick fits. You connect it to the assistant you already use and say "remember this"; later, when it's relevant, that assistant brings it back. One number gets measured.

When this isn't for you

Three cases, honestly:

If you already work in Notion, or something like it. Your material is there and your team is there, and adding a second store is usually a net loss. Connect an assistant to that instead.

If you enjoy organizing. If splitting things into folders and tagging them is satisfying, or you need that structure before you can settle, we have nothing to offer you — there are no folders and no tags in our store. It was built to make that unnecessary.

If you use one AI and don't plan to switch. Built-in memory keeps getting broader. If you're not moving between assistants, the reason to keep this outside is weak — that reason usually turns up on the day you switch.

What this doesn't do

To be clear about the edges:

If you want to try it, connecting takes about a minute:

← All posts