Author: abavyka
Projects and project memory
Projects in claude.ai have three layers
• Instructions and files — the stuff you put in by hand. Static, gets mixed into every chat in the project
• Project memory — what Claude remembers on its own from the chats in this project: who you are, what you’re working on, what you decided, how you like your answers. Saved as separate topics right during the conversation
• Search across past chats.
Important: each project has its own separate memory. What you discussed in the flat-rental project won’t leak into the work project, and vice versa. Chat search also only works within the project.
It’s memory that lets you spin up a new chat for every task and not lose the thread. No need to drag one giant chat along for a week: the context is fresh, and Claude remembers the important stuff from past conversations anyway.
Gotchas
• Memory is a condensed summary, not a transcript. Details can get lost, and some stuff goes stale
• You can steer it right from the chat: “remember this” or “forget that”. Anything critical is safer to put in the project instructions
• Memory eats context too, like everything else
• For a chat where you don’t need memory, you can turn it off: that chat doesn’t read or add to memory
The Claude Code counterpart is auto memory: Claude keeps its own notes about the project (build commands, patterns, mistakes) and loads them at the start of a session. You can view and edit them via /memory.
Read more:
• Claude projects and memory
• Auto memory
• https://support.claude.com/en/articles/11817273