Super Memory

Persistent shared knowledge layer for AI coding agents.

AI coding agents (Claude Code, Cursor, Aider) suffer from session amnesia. Every new session starts cold — rediscovering architecture, re-exploring codebases, repeating mistakes. Super Memory fixes this with a .memory/ directory that any agent can read and write.

How It Works

A .memory/ directory in your repo holds structured markdown files with YAML frontmatter:

• systems/ — Architecture docs for each subsystem (committed)
• decisions/ — Architectural Decision Records (committed)
• gotchas/ — Known pitfalls and bug patterns (committed)
• research/ — Investigation findings (committed)
• people/ — Per-person context and ownership (committed)
• sessions/ — Session summaries (gitignored)
• scratch/ — Working memory during sessions (gitignored)

Agents read these on session start (warmup), write discoveries during work, and commit knowledge at session end (cooldown).

Quick Start

# Install
pip install atlas-memory

# Initialize in your project
cd your-project
atlasmemory init

# Set your identity
atlasmemory whoami yourname

# Auto-discover codebase systems
atlasmemory systems discover

# Start coding with your AI agent — it reads .memory/ automatically

Non-Interactive Init

atlasmemory init \
  --project "my-project" \
  --me mitch \
  --platform claude-code \
  --tracker beads \
  --obsidian \
  --auto-discover

CLI Commands

atlasmemory init                              # Initialize .memory/
atlasmemory whoami [name]                     # Get/set identity
atlasmemory systems                           # List system docs
atlasmemory systems discover                  # Auto-discover modules
atlasmemory systems stale                     # Show stale docs
atlasmemory query --system auth               # Query by system tag
atlasmemory query --concern race-condition    # Query by concern tag
atlasmemory summary                           # Regenerate SUMMARY.md
atlasmemory health                            # Check memory health
atlasmemory check-conflict path/to/file.py    # Check DO NOT TOUCH claims

Agent Platforms

Super Memory integrates with:

Obsidian Integration

Pass --obsidian to atlasmemory init to generate Obsidian vault config. Then open .memory/ as an Obsidian vault for graph view, tag filtering, backlinks, and Dataview queries.

Issue Tracker Integration

Super Memory complements (doesn't replace) issue trackers. Use --tracker with: beads, linear, github, or jira to enable backlink fields in templates.

Team Usage

1. Each team member runs atlasmemory whoami theirname
2. Memory files are committed to git and shared
3. Person files track who's working on what (agents check "DO NOT TOUCH" sections)
4. Sessions and scratch stay local (gitignored)

License

MIT