Does Claude remember?

Yes — Anthropic added memory in 2025, so Claude can now carry context across chats. But it's Anthropic's memory: it lives in their cloud and can't follow you to another model. Here's how Claude's memory really works — and how to make it yours.

How Claude handles memory today

Inside a single conversation, Claude remembers what you've said. That's the context window — the running transcript of the current chat. Ask a follow-up and Claude has the earlier turns to work from. So far, so good.

Across conversations, the picture changed in 2025. Anthropic added memory to Claude — it can search your past chats, carry context forward, and keep separate memories per Project — and that memory reached free users in March 2026. So the old answer, that Claude forgets the moment you open a new chat, is no longer accurate. You don't always have to paste last week's project back in for Claude to pick up where you left off.

Which moves the real question. It isn't whether Claude remembers anymore — it's whose memory that is, and whether you can take it with you. Claude's memory lives in Anthropic's cloud, under Anthropic's retention and training terms. You can't open it as an editable file on your own machine, and it can't move to GPT or any other model — it's tied to your Claude account and a Claude subscription. That's a different limitation than forgetting, and it's the one this page is about.

Claude Projects: useful, but locked in

Claude Projects are the most structured piece of this. A Project is a workspace where you set shared instructions and add reference files, and every chat you start inside it draws on that same context. For a long-running piece of work — a codebase, a book, a client — that's genuinely useful. The chats inside a Project stop starting from zero, because the Project holds the background for them.

Credit where it's due: this is good context scoping. If your work fits neatly into a Project and you live inside Anthropic, it solves a real part of the problem. But it's worth being precise about what a Project is and isn't.

• It's bounded by the Project. Context is shared among the chats inside that one workspace. It isn't a memory of you that spans everything you do with Claude.
• It's locked to Anthropic. A Project lives in your Claude account. There's no version of it that GPT or any other provider can read. The day you want to try a different model, the Project doesn't come with you.
• It isn't a file you own. You can't open the accumulated context as an editable document on your own machine, correct what's wrong, and keep it. It's context the vendor holds for you, not memory you hold yourself.

So Projects are a good answer to "how do I scope context inside Claude" and a poor answer to "how do I own a memory that follows me." Those are different questions, and the gap between them is the whole reason this page exists.

What you actually want from AI memory

Strip away the product names and the thing most people are reaching for is fairly specific. Not "a smarter chatbot" — a memory that behaves like a memory. Concretely, four properties:

• It persists across chats. Not just within one Project or one thread — across every conversation, so you explain a thing once and it stays explained.
• You can edit it. When the memory gets something wrong, you open it and fix it. The record of you is yours to correct, not a guess locked behind an interface.
• It survives a switch. Move from Claude to GPT — or to a local model running on your own machine — and the memory comes too. The provider is the part you swap; the memory is the part you keep.
• It's private, and never trained on. It lives on your machine, not in a vendor's cloud, and it isn't quietly improving someone else's model.

Claude's memory now delivers the first — it persists across chats. But not the other three: you can't freely edit it as a file, it doesn't survive a switch to another model, and it lives in Anthropic's cloud rather than on your machine. Projects add structure to that first property without changing the other three. That's the shape of the gap — and what a memory layer you own is for.

How Memrith gives Claude a portable memory

Memrith is a desktop app that sits between you and Claude and gives Claude the memory it doesn't keep on its own. You bring your own Anthropic key — Memrith is BYOK, so the calls go directly from your machine to Anthropic under your own account, and Memrith never sees your conversations. There's no Memrith markup on AI usage; you pay Anthropic for what you use.

The memory itself is a local, editable file on your machine. As you work, Memrith keeps a model of your projects, decisions, and the way you think, and supplies the relevant parts to Claude at the start of each conversation. So Claude answers in context — it knows the arc of the work, not just the prompt you just typed — without you pasting last week's history back in.

And because the memory is yours rather than Anthropic's, it isn't tied to Claude. Decide GPT is better for what you're doing this year, or route through OpenRouter, and switching is a Settings change: paste the new key, save, keep going. Your memory is unchanged. The next chat opens with the same context, routed through the new model. That's the part Claude Projects structurally can't do — your memory outliving the provider you happened to start with.

Claude Projects vs Memrith

The honest trade-off

Here's the part a landing page usually skips. If all you want is for memory to work the moment you open the app, with nothing to set up, Claude's first-party features are simpler today. There's no key to paste and no extra app to run — you stay inside one product. That's a real advantage, and pretending otherwise would be dishonest.

Memrith asks for a little more at the start. Because it's BYOK, you set up an Anthropic key first — a few minutes, with a walkthrough for each provider. The reason to accept that is everything above: a memory that spans every chat, that you can edit, that you own as a local file, and that comes with you the day you switch away from Claude. You're trading a little setup now for portability and ownership later.

If you'd rather not switch providers and don't mind that the memory stays Anthropic's, Claude Projects are a reasonable place to stop. Choose Memrith when you'd rather own the memory you spend years building — and keep it whichever AI you're using when you look up in three years. More on that case in why Memrith is BYOK and own your memory. And if you're weighing it against a notes app, it isn't a second brain — that's a different job.

Common questions