What to look for in an AI chat app with memory
Most apps that advertise memory are offering searchable history. Real memory is different: it is applied automatically in a new conversation, it is visible and editable, it can be deleted entry by entry, and it can be switched off. Ask for a settings page, not a marketing page.
· 4 min read
Every assistant now says it remembers you. The word covers at least four different things, and only one of them changes how the product feels day to day.
The four things "memory" can mean
- History: a list of past conversations you can search. Useful, but you have to do the retrieving.
- Within-thread context: the model can see the whole current conversation. Every chat app has this.
- Extracted memory: a store of facts and context distilled from your conversations and applied automatically.
- Manual memory: a place where you write down what the assistant should always know.
The last two are what people mean when they say an app "knows" them. The first two are table stakes.
Five questions that separate real memory from a transcript
- If I start a new chat and say nothing, does it already know my project?
- Can I see the list of things it believes about me?
- Can I fix one wrong entry without wiping everything?
- Can I turn memory off for a sensitive conversation?
- Does memory follow me across devices, or does it live in one browser?
Why editability is the whole ballgame
An assistant that remembers the wrong thing is more dangerous than one that remembers nothing, because it applies the error confidently and repeatedly. Memory you cannot inspect is a rumour, not a feature. The minimum viable interface is a list, an edit, a delete, and an off switch.
Model choice matters more once memory exists
Without memory, a chat app is a thin shell around one model, and switching apps is cheap. With memory, switching costs real context, so the rest of the product has to earn the lock-in. That means more than one model — a fast tier for small questions and a heavy tier for hard ones — and an escape hatch: the ability to export or delete what it knows.
Where Swirl lands
Swirl implements extracted memory plus manual entries, exposes both in a Memory settings tab, and lets you delete individual memories or disable the feature. It routes across several frontier models, keeps artifacts editable, and connects to outside tools over MCP. If you are evaluating alternatives, ask each one the five questions above and compare the answers, not the landing pages.
Questions
Do free AI chat apps have memory?
Some do, in the limited sense of searchable history. Persistent, editable memory is usually a paid feature because it requires storage and retrieval infrastructure per account.
Is persistent memory a privacy risk?
It can be. The mitigations are visibility, selective deletion, and an off switch. A product with no memory settings page is asking you to trust something you cannot audit.
Does memory improve with more conversations?
It should, up to a point. More conversations give the system more signal, but good retrieval matters more than volume — a large store that surfaces the wrong facts is worse than a small accurate one.
Keep reading
· 5 min read
What is an AI chat app with memory?
An AI chat app with memory keeps context between conversations instead of starting from zero. Here is what that changes in practice, and what to check before you rely on it.
· 5 min read
How to choose an AI chat app
Six criteria that actually predict whether you will still be using an AI chat app in a month, and how to test each one in an afternoon.
· 6 min read
Swirl vs ChatGPT: which one fits how you work?
A practical comparison of two approaches to AI chat: one assistant from one vendor, or several models behind one memory. No benchmarks, just the trade-offs that show up in daily use.