Memory Bank

Features

• Token-efficient persistent memory system
• Layered architecture (Global, Project, Session)
• Branch-aware context management
• Smart compression and archival of memory
• Session diffing and conflict detection
• Recovery mode for memory reconstruction
• Session continuation protocol for seamless resumption
• Context budget tracking and warnings

Use Cases

• Maintaining context across multiple AI coding sessions without re-explaining
• Reducing token consumption on session warm-up by loading compact memory
• Switching between different feature branches with distinct project contexts
• Generating developer handoff documents from project memory

Non-Goals

• Storing or managing secrets or credentials
• Executing arbitrary code or commands
• Acting as a general-purpose file manager or system tool

Workflow

1. Detect memory files and git context at session start
2. Perform session diff to identify changes since last memory update
3. Score memory health and greet user with summary and next step
4. Update memory mid-session on explicit user commands or task completion
5. Perform full memory write to MEMORY.md on session end
6. Compress memory if it exceeds defined thresholds