Memory in 📝Claude Code is the file-based persistence system at ~/.claude/projects/<project>/memory/, indexed by MEMORY.md, that survives across conversations and informs future sessions automatically.
The memory system stores discrete facts as individual markdown files alongside a MEMORY.md index that ships into each new session's system prompt. Files are categorized by type — user_*.md captures user preferences, feedback_*.md captures corrections and confirmed approaches, project_*.md captures ongoing initiatives, and reference_*.md captures pointers to external systems. The index keeps each entry to one line so the loaded context stays under a hard ~150-line cap.
Claude writes memories via the /remember slash command or autonomously when it detects a durable, non-obvious fact worth preserving. Reads happen automatically — every session loads MEMORY.md, then opens individual memo files on demand when their descriptions match the conversation. The judgment threshold is intentionally high: ephemeral conversation state, code patterns derivable from the codebase, and anything documented in CLAUDE.md should not be saved.
For practitioners with richer knowledge systems, memory is often shadowed by an external library. Brian's setup forwards every memory to 📝MythOS via the 📝MythOS Claude Code Skill, making the local file-based memory a hot cache and MythOS the canonical durable store accessible from any device.