Claude Memory Explained: How Claude Remembers You Across Chats

The 4 layers, nearest to most durable

• Conversation Memory — the current chat itself. Nearest, fastest, needs no setup — the memory everyone already uses without thinking about it.
• System Instructions — Settings → General → 'Instructions for Claude.' Standing preferences applied to every new chat outside a Project.
• Reflect (manual opt-in) — Settings → Memory → 'Generate memory from chats.' Once on, Claude synthesizes a running summary of who you are from your non-Project chats, refreshed roughly every 24 hours. Available on Free, Pro and Max — not Team/Enterprise.
• Project Memory — lives entirely separate from general chat: its own instructions, its own files. Project chats never feed Reflect — deliberately excluded from the general memory synthesis, which is exactly why a client- or case-specific Project stays isolated from your general 'who is this person' summary.

What Cowork and Claude Code actually inherit

Cowork and Claude Code do not draw on your Claude Chat Reflect memory. Only your standing system instructions flow into them — the chat-history summaries stay behind in Claude Chat. Worth confirming against your own workspace before treating this as permanently settled, since these are two of Anthropic's newest, fastest-moving products.

Why this actually matters

Understanding these four layers is what lets you stop re-explaining yourself constantly — a standing fact goes in System Instructions once; a client-specific detail goes in that client's Project; everything else, Reflect quietly picks up on its own. Getting this right is also directly a token-saving habit, not just a convenience one.