The first time Claude remembered something about me without being told again, it was a small thrill. It opened a new chat and referenced a project I had mentioned days earlier. And then, a few minutes later, it became obvious how little it actually held onto. It knew a preference. It did not know my work.
This gap is the topic: how do you inspect and maintain the context Claude keeps between conversations? Built-in memory has developed further. An external store can help some workflows, but is not required for Claude to know your work.
This is the tenth post in a beginner's series on Claude. The first one mapped out the whole tool. Here we get to the piece that, more than any model upgrade, decides whether Claude feels like a clever stranger or a colleague who knows you.
What the Built-In Memory Actually Does#
Memory is available and on by default on Free, Pro and Max. Team and Enterprise may need workspace-owner approval. You can inspect and edit saved memories in Settings → Memory. Projects have their own context and separate memories; search of previous chats is an additional paid feature. Check the references shown for an answer and ask when you need to know which memories it used.
For a lot of everyday use, that is genuinely enough. It stops you re-introducing yourself every morning. It is a real improvement over the years when every chat started cold.
Where It Stops#
Built-in memory can also retain project and work information. You can edit entries and transfer memories using Anthropic’s import/export feature. It is not an inaccessible store. Limits remain: not every decision is captured, export is not live synchronization and import can be incomplete. Authoritative decisions need a maintained source.
For reliable retrieval of decision reasons, client details or rejected approaches, maintain a short decision log or suitable project files. Ask Claude to record important entries with a date and source, then check them. Native memory can help, but should not hold their only copy.
What a Real Memory Looks Like#
An optional next step is a store that several compatible clients can read and write through connectors. This helps when you need the same maintained context across providers. Check sign-in, permissions, export, deletion and retrieval quality. A connected server does not automatically follow you into every tool; each client must support and actually use it. The next post covers setup.
When it is in place, something shifts. The tool stops being a goldfish that greets you fresh every morning and starts being the colleague who has been with you for years and does not need the backstory. That shift is not a technical detail. It is the moment memory stops being a feature you toggle and becomes the reason the whole thing is worth using.
The Honest Catch#
Memory helps only when retrieved information is correct and relevant. Old notes can pull an answer in the wrong direction. For important tasks, compare a run with memories against one with current source material. Check not just whether something was found, but whether its source, date and decision still apply.
That is worth sitting with before you go chasing memory features. A pile of everything you ever said is not memory, it is a junk drawer. Real memory is selective. It keeps what will matter later and lets go of what will not, the same way a good assistant remembers the decision and forgets the small talk.
Why This Beats a Bigger Model#
Good context can help recurring work more than switching to the next model. This is not a universal ranking: a stronger model can make a real difference on difficult tasks. Test with your material whether maintained memories reduce errors and save retrieval work. Context and model capability work together.
Where to Start This Week#
Start by actually using the built-in memory well. Open the memory settings, see what Claude has kept, and correct it. Tell it the handful of things about your work you are tired of repeating and let it hold them. That alone will show you how much better it feels when the tool knows even a little about you. Then, when you outgrow the light version, the next post shows you how to connect the deeper kind.
The industry keeps selling the model as the thing that matters. Spend a week paying attention to memory instead, and you will see where the real leverage was hiding. If you want a structured path through all of it, our free StudioMeyer Academy covers memory in depth. Next in the series, the connectors that make this possible, and the standard behind them.
Sources and product status#
Claude memory, chat search and Incognito / Import and export Claude memory
