claude-mem auto-captures your Claude Code sessions through hooks, AI-compresses them into summaries, and injects those summaries as context when your agent starts up. Brain OS captures structured operational state (decisions, plans, blockers, focus) through typed MCP tools. Different categories of AI memory, designed for different problems.
claude-mem is the incumbent in the AI-memory space. Hooks auto-record your sessions in the background, AI compresses them into summaries, and the next time your agent starts working, those summaries get injected as context. You don't write to claude-mem; it records you. The mental model is an AI-compressed transcript log; the user is passive.
It works well for one specific job: continuity of dialogue. If you want your agent to remember what you said three weeks ago, or carry tone and references across sessions, transcript-based recall is the right tool. claude-mem is the most mature option in that category.
Brain OS does not touch transcripts. It does not auto-capture anything. Instead, your agent writes typed, structured operational state about your project: decisions with reasons and rejected alternatives, plans with phases, blockers, focus, momentum, and recurring patterns. It reads back through typed MCP tools (focus_get, decision_log, plan_read, pattern_detect, semantic_recall), not by searching old messages.
The difference: claude-mem answers "what was said?" Brain OS answers "what is true about the work?"
.brain/ shared by Claude Code, Cursor, Zed, GitHub Copilot, any MCP clientdecision_check returns conflict/caution/clear with prior reasoningyou want your agent to recall what was said in earlier sessions, surface a snippet you discussed, carry conversational tone, or replay reasoning you don't want to redo.
you want your agent to operate from your project's current state: know what you've decided and why, what's blocked, and what matters this week. No re-explaining or scrolling chat history.
The two are complementary, not competing. They sit on different axes. Many projects benefit from both: claude-mem for conversation recall, Brain OS for operational truth.