Memory System

MiMo-inspired, SQLite-backed memory store.

Core concepts

• memories table — importance, confidence, access_count, type ranking
• memory_timeline — event lifecycle tracking
• Budgeted injection — importance x recency x confidence, fits token budget
• Auto-init + legacy migration — first-use setup is automatic

Memory hierarchy

.clew/memory/
  MEMORY.md       # Project facts and context
  DECISIONS.md    # Architecture decisions
  TASTE.md        # Coding style preferences

Commands

Command	Description
/memory init	Bootstrap project memory
/memory scan	Detect stack, language, entrypoints
/memory rebuild	Reconstruct context from ranked memories
/memory recall	Recall memories ranked by score
/memory feedback	7 signals: accepted, rejected, corrected, preferred, etc.
/memory dashboard	Unified view of MemoryDB, Dream, Distill, Peer Sync
/memory search	Search stored entries

Auto-consolidation

• Dream (7-day) — consolidates recent memories
• Distill (30-day) — long-term pattern extraction
• In-compact extraction — saves structured facts during context compaction
• Peer memory sync — imports memories from all connected peers

Cross-session

Memory feedback signals update importance/confidence scores. The preferred signal writes to TASTE.md. Auto-memory extraction during /compact saves [decision], [architecture], [taste], [bug] facts.