Claude has one memory now 🧠

Memory now carries across chat and Cowork — hand Cowork a task and the context you built up in chat is already there, and what happens in Cowork comes back. It is on by default, so this step is usually confirming rather than enabling. The part almost nobody does: open Topics and read the short files Claude has saved about you.

WHERE: Settings, then Memory. Confirm the toggles, decide the sensitive-topics one deliberately, then scroll to Topics and read three of them. Every file is readable, editable and deletable.

Memory picks things up as you go, but the fastest way is to just tell it everything at the start, deliberately, one time.

Paste this and fill it in:

Remember this about my current project so you have it in future chats. Project: [what it is]. Goal: [what done looks like]. Stack or tools: [what you're using]. What I've already tried: [the dead ends, so you don't suggest them again]. Constraints: [budget, time, skill, anything fixed]. How I want you to help: [review my work / write drafts / ask me questions / all of it].

A Project is a folder with its own instructions and its own files. Everything in it applies to every chat inside it, and only inside it. Job hunt, side business, a class, planning a trip. Files go in once.

WHERE: Projects in the sidebar, then New Project. Put the instructions in the Project, not your global profile, so they don't leak into unrelated chats.

Open a brand new chat and don't explain anything. Just ask something that needs context.

Try this in a fresh chat:

Without me explaining anything: what am I currently working on, what have I already tried, and what would you suggest I do next? If you don't actually know, say so instead of guessing.

When a project ends, tell it. Otherwise it keeps giving you advice for something you finished three months ago, which feels exactly like the model getting worse.

Every month or so:

List everything you remember about me and my projects. Flag anything that looks out of date. I'll tell you what to drop.