Memory and Projects for beginners: what they are and when to switch them on

Every chat you open starts with no memory of you, and once you understand why, the two fixes take about ten minutes to set up.

You open a fresh chat, ask a great question, get a sharp answer. The next morning you open another chat and the model has no idea who you are. No recollection of what you're building, what you told it yesterday, or how you like your answers formatted. You re-explain yourself. Again. This is the single most common frustration CEOs have with AI, and it has a simple fix.

Why a fresh chat forgets you

Every chat runs inside a context window: the model's working memory for that one conversation. Close the tab, start a new chat, and that working memory is empty. The model does not have a long-term memory the way you do. It has a window, and the window resets.

Think of it like a desk. Whatever is on the desk, the model can use. Whatever isn't there, it fills in by guessing. When you start a fresh chat, the desk is bare. You are not on it. Your company is not on it. Nothing you said last week is on it.

This is not a flaw. It is how the architecture works. But it does mean that if you want the model to know anything about you beyond what you type today, you have to supply that context deliberately. Two features do that job.

The two fixes

Projects: a workspace for one recurring use case

A Project is a scoped workspace for a specific job. You upload the relevant documents once (your pricing sheet, your brand guide, your top ten customer emails), write a few sentences of standing instructions ("answer in plain English, never invent a number I didn't give you"), and every chat started inside that Project inherits all of it.

The project doesn't know anything about you personally. It knows about the job. That's the point. A Project for board prep has your board deck and your last six months of metrics. A Project for a specific client has the contract and the account notes. A Project for monthly finance review has the chart of accounts and your reporting template.

Free accounts get a few Projects. Paid accounts are effectively unlimited. The full how-to on setting one up is at What is a Claude Project?.

Memory: account-level facts about you, everywhere

Memory is different. Where a Project is a workspace for one use case, Memory is a running profile about you that travels into every chat, regardless of which Project you're in or whether you're in a Project at all.

It holds things like: your role, how you like output formatted, ongoing context about your company, names you keep referencing. You can open Settings, read exactly what the model has saved, edit anything that's wrong, and delete anything you don't want there.

If you ever want a conversation with no trace, there's a temporary chat mode (sometimes called Incognito) that runs with Memory switched off for that session.

Memory is now broadly available across tiers. You don't need to configure it. It builds passively from your normal chats.

One honest note: if most of your chatting happens inside Projects, Memory builds more slowly because Projects have their own context. Heavy Project users sometimes find their Memory profile is thinner than they expected.

The distinction in one line

Memory is about YOU, everywhere. A Project is a workspace for one use case. They are complementary, not competing. A well-built Project combined with a clean Memory profile is noticeably better than either one alone.

When to turn each on

You do not need to set both up in the same afternoon. Here is the beginner sequence.

Week one: one Project, one job.

Pick your single most repetitive AI task. Board prep. Drafting client replies. Finance review. Create one Project for it. Add two or three of the most relevant documents. Write three sentences of instructions. That is the whole starting move. Don't try to build five Projects before you've seen how one works.

Leave Memory on. Review it in a couple of weeks.

Memory is passive by default. You don't need to do anything. After a few weeks of normal use, open Settings and read what it saved. Delete anything wrong or outdated. If it captured something useful you didn't realize it logged, that's the product working.

Privacy check before you upload anything.

Anything you put into a Project is processed by the AI. Do not upload contracts, personnel files, or anything sensitive without checking your account's data settings and your company's AI policy first. Is your data safe in AI? is worth ten minutes before you load the serious stuff.

Where to go from here

If you haven't spent thirty minutes in Claude just getting comfortable before setting up Projects or Memory, start at your first 30 minutes with Claude. It's a faster path to the second thing than figuring out the second thing first.

Once the basics feel natural, Claude Memory goes deeper on what Memory tracks, how to shape it deliberately, and when power users override it for specific conversations.

A Project takes ten minutes to build. A decent Memory profile takes a few weeks of normal use. Between those two timelines, Memory is the lower-effort move right now. Set up one Project this week and let Memory do its thing in the background. That combination compounds fast.