Lesson Content

The default is amnesia.

Every Claude session starts fresh. When you close a conversation, the context of that conversation – what you told Claude about yourself, your project, your preferences, your constraints – does not automatically carry into the next session. Claude does not remember that you told it your name last week, or that you prefer bullet points over paragraphs, or that you are working on a specific project with specific requirements.

This is not a bug. It is a deliberate design choice rooted in privacy and predictability. But for users who interact with Claude regularly, it creates constant re-establishment overhead – re-explaining the same context over and over at the start of every session.

The memory system exists to eliminate that overhead. But it requires intentional setup.

What persists by default.

These things survive between Claude sessions automatically:

• Conversation history: Past conversations are saved in your account and accessible in the Recents sidebar. You can re-open them and continue. But they do not carry context automatically into new conversations.
• Projects and their contents: Files uploaded to a Project and standing instructions written for a Project persist and are active whenever you open a conversation in that Project.
• Custom Instructions: Profile-level instructions you write in Settings apply to every new conversation automatically.
• Account settings: Your subscription, profile, payment information, and preferences are account-level and always present.

What disappears by default.

These things do not carry across sessions automatically:

• The context you established in a conversation (names, preferences, constraints, background you explained)
• Any instructions you gave Claude within a conversation that were not in Custom Instructions or a Project
• The working knowledge Claude built up about your project during a session

The five-layer Claude memory stack.

A fully configured Claude memory setup uses five layers, each serving a different purpose:

1. Custom Instructions (cloud-stored, account-level): Apply to every conversation in all modes. Your standing brief – who you are, how you work, what you prefer.

1. Projects (cloud-stored, account-level): Persistent workspaces with uploaded files and standing instructions for specific areas of work. Active whenever you open a conversation in the Project.

1. CLAUDE.md files (locally-stored, directory-level): Instruction files Claude Code reads at the start of every Code session in a given directory. Project-specific persistent context for Code mode work.

1. Auto-memory files (locally-stored, account-level): Markdown files in your Claude configuration directory that Claude Code writes to and reads from across sessions. Store facts about you, your preferences, ongoing projects, and external references.

1. Memory feature (cloud-stored, conversation-level): Claude's built-in memory feature that retains facts across Chat conversations. Configurable in account settings. Verify current availability and controls in your Claude version.

Cloud vs local.

Custom Instructions, Projects, and the memory feature are stored in your Claude account on Anthropic's servers. They are accessible from any device where you sign in.

CLAUDE.md files and auto-memory files are stored on your local machine. They are accessible only on the device where they exist unless you sync them manually or via a backup system.

Understanding this distinction matters for backup planning (Lesson 6) and for device transitions.