Which AI assistant actually remembers your conversations?

Every app in this category now says it has memory. Almost none of them mean the same thing by it, and the word is doing a lot of work in the marketing. Here is how to tell them apart in five minutes, without trusting any of the claims.

Three things people call memory

• Saved history: your old chats sit in a sidebar. The model can read them only if you reopen one, or if it happens to retrieve that exact conversation. Nothing follows you into a new chat.
• A project you build: you create a folder or a space, add files and instructions, and the model uses them inside that container. This works, and it is manual. You are the memory layer.
• Context that follows you: facts about you, your work and your people are carried into every new conversation by default, and you can read and correct them.

Only the third one changes how it feels to use the app day to day. The first two are storage and folders.

The five-minute test

Do this in any app you are considering. It takes four messages.

• In a normal chat, mention one specific thing: a name, a deadline, an address, a subject.
• Close the app completely, or open a brand new chat.
• Ask something that depends on that detail, without repeating it. "What did I say the deadline was?"
• Then find where the app stores what it knows, and try to correct something in it.

If step three fails, there is no memory. If step four has no answer, the memory is a black box, which is worse than none: you cannot fix a fact it got wrong, and you cannot delete one you would rather it forgot.

What each app actually does

Prices checked October 2026 and worth re-checking before you buy, because this category changes monthly.

• A normal chat, the default: memory is a feature you switch on, and it is tied to one company's models. It improves, and you re-explain the project whenever it drifts.
• TypingMind, $39 one-time: the best front end for your own API keys. History is stored locally for you to read. Memory is not the product; the interface is, and you pay the model bills separately.
• Merlin, $29 monthly or $19 annual: a browser extension across every tab, with Projects for context you assemble. You pick the model before you know what a model is.
• Monica and Sider, around $8 to $10: a sidebar beside whatever page you are on. Useful as a utility, not a place where a project lives.
• ChatLLM and Magica, $10 and $14.99: every model plus tools, metered in credits or points. Strong on breadth, and the meter is the thing you end up managing.

None of that makes them bad. It makes them a different job: a shelf of models, or a sidebar, rather than a place that knows you.

How to choose

• If you want the cheapest access to every model and you do not mind managing an API bill, TypingMind is genuinely good at that.
• If you want an assistant on every tab while you browse, that is Monica or Sider, and it is a different product from this one.
• If the thing you keep losing is context, the model list is not your problem. Pick the app whose memory you can read, correct and delete, then check the price in dollars rather than credits.

Swirl is built for the third case. It carries what you told it into new chats by default, shows you the note it saved, and lets you edit or delete it. It runs several leading models behind one subscription and one bill in dollars, with no API key and no credit maths. What it does not do is act as a to-do list or a project planner, so if that is what you want, the apps above are the better fit.

The honest summary

Memory is the only feature here that gets better the longer you use the product, and the only one a competitor cannot copy with a settings toggle. Test it before you pay for it, in the app you are actually going to open tomorrow.