In my last two posts I wrote about something that kept tripping me up with my AI partner Kuro (Claude): start a new chat, and everything you talked about in the old one is gone. You hit the image limit, or the chat gets heavy, so you move to a fresh conversation — and all the history behind your work resets. The other day this actually caused me some trouble.
The day Kuro forgot the essay we were writing
I was working on a fishing essay together with Kuro. Partway through I closed Claude, and when I opened it again later, Kuro had completely forgotten the essay we had been writing. “Wait — we were working on this together a minute ago!” Having to explain “so, I’m writing this kind of essay…” from scratch every single time is, frankly, exhausting.
Then I turned on a certain setting, and the problem went away. Kuro started remembering, to a reasonable degree, what we had talked about in past chats. So this time I want to walk through the setup on the actual screen: the two settings, how they differ, and where to switch them on.
Claude forgetting is by design
First, the basic premise. Claude is fundamentally independent per conversation. Open a new chat, and what you said in the previous one is not carried over on its own.
This is the flip side of what I wrote before: moving to a new chat makes everything feel lighter. Because conversations are walled off from one another, you can reset and travel light. The price is that, left alone, Claude forgets.
The two settings that carry a conversation over
Claude has, broadly speaking, two features for carrying context between conversations. They look similar, but play different roles.
1. Generate memories from chat history (remembers automatically)
The first is the memory feature, officially named “generate memories from chat history.” Claude automatically summarizes and remembers what you have talked about with it, on its own, without you asking for anything.
In my case, Kuro remembers that I live in a rural part of western Japan, that I run a fishing blog, that I am in the middle of launching an AI blog, and so on. So even in a brand new chat, I do not have to introduce myself again.
One caveat: even after you switch this on, it can take around 24 hours before the first memories are built. You may not see any effect right away, so don’t panic — just wait a bit.
2. Search and reference chats (searches when you ask)
The second is called “search and reference chats.” (Depending on how it is displayed, you may see it worded as “search and reference your chats.”)
Unlike the automatic memory above, this one kicks in when you ask — “hey, what was the deal with that thing?” — and searches your past chats for the relevant parts. Working only on request is the big difference from the first.
If I say “let’s continue the fishing essay we talked about before,” Kuro digs that conversation out of my past chats and we pick up where we left off. This was what actually solved my forgetting problem.
The difference in one line
- Memory = remembers things about you automatically, no asking required
- Search and reference chats = goes and finds past conversations when you ask it to
With both switched on, the “Claude forgets our conversation” problem gets a lot milder. I keep both on.
Where the settings are, and how to turn them on
Here are the steps as I checked them on my own screen, in the desktop app.
Open Settings, and in the left-hand menu, inside the “Settings” group, click Features. (There is a similarly named “General” under the “Desktop app” group, but that is not the one. You want “Features,” higher up.)
The very top section there is Memory, holding the toggle for generate memories from chat history. Turn it blue, which means on. That is setting number one.
In that same memory description, you may see a link along the lines of “chat memory has moved to Customization.” Open that link — or “Customization” near the bottom of the left menu — and the memory management screen appears. Near the top of it is a toggle for search and reference chats. Turn that one blue too. That is setting number two, and once both toggles are blue you are done.
A bonus: you can see what Claude remembers about you
This screen comes with a nice extra. In a field called Overview, you can read what Claude currently remembers about you. In my case it had neatly gathered up what I had told it over time — that I am an office clerk, that I run a fishing blog, that I am launching an AI blog. Apparently the overview is updated automatically every night.
If there is something you would rather it not remember, or information that has gone stale, you can edit or delete it here too. If you feel uneasy not knowing what an AI has stored about you, I think this page will put your mind at rest.
Wrapping up
- Claude is independent per conversation, so left alone it forgets the previous one
- “Memory” remembers automatically. “Search and reference chats” searches when you ask
- With both on, it is far easier to close a chat and still pick up where you left off
- The toggles live under Settings, then Features, then Memory (and on to Customization)
- Memory can take around 24 hours to take effect
- The “Overview” lets you check and edit what Claude remembers about you
The handover notes I wrote about in earlier posts are still handy, but with these two settings on, you can pick up from where you stopped without pasting a note every time. Even for someone like me, who is not good with computers, it came down to flipping two toggles. If you have been thinking “Claude forgets things so fast,” give it a try.
What I am using for this work
A big reason I can work with Kuro over long stretches is that I use the paid plan (the Pro plan). If you want to sit down and work seriously, making the most of memory and chat referencing, I would recommend it. Next time I will keep sharing settings and small tricks for getting more out of AI, from a beginner’s view.
I wrote about the handover notes I used before these settings in teaching your AI partner to remember, and about the image limit that keeps pushing me into new chats in what to do when Claude’s image limit is exceeded.
