Blitz

Memory System

Legion CLI includes a persistent memory system that loads project-specific context and session learnings to improve AI responses.

Overview

The memory system provides:

• LEGION.md loading from project root and global config
• Session memory persistence across conversations
• Automatic injection into system prompts
• Structured storage in JSON format

How It Works

┌─────────────────────────────────────┐
│         Memory Loader               │
├─────────────────────────────────────┤
│  1. Load LEGION.md (project)        │
│  2. Load LEGION.md (global)         │
│  3. Load session memory             │
│  4. Format context                  │
│  5. Inject into system prompt       │
└─────────────────────────────────────┘

LEGION.md

Create a LEGION.md file in your project root to provide persistent context:

# Project Memory

## Architecture
- Frontend: React with TypeScript
- Backend: Node.js with Express
- Database: PostgreSQL with Prisma

## Conventions
- Use functional components with hooks
- All API endpoints follow RESTful patterns
- Error handling uses custom error classes

## Decisions
- 2024-01-15: Switched from MongoDB to PostgreSQL
- 2024-02-20: Adopted Zustand for state management

## Learnings
- The auth module requires special handling for JWT refresh tokens
- Database migrations must be run in order

Global LEGION.md

For preferences across all projects, create ~/.legion/LEGION.md:

# Global Preferences

## Coding Style
- Prefer functional programming
- Use async/await over callbacks
- Always handle errors explicitly

## Tools
- Use ripgrep for searching
- Prefer fd over find
- Use bat for file viewing

Session Memory

Session memory stores learnings from conversations:

// Memory entries are stored in ~/.legion/legion-memory.json
{
  "key": "learning-1234567890-abc123",
  "content": "The auth module requires JWT refresh token handling",
  "source": "auto",
  "timestamp": 1710000000000
}

Memory Sources

Source	Description
legion.md	LEGION.md file content
user	User-created memory entries
auto	Automatically saved learnings

Automatic Context Injection

Memory is automatically loaded and injected into the system prompt:

# Project Memory
The following memory has been loaded from LEGION.md and session memory.
Use this context to inform your responses about project conventions...

# LEGION.md content
[Your LEGION.md content here]

# Session Memory
## learning-1234567890-abc123
The auth module requires JWT refresh token handling

Using Memory

In Conversations

Memory is automatically available. The AI will use it to:

• Follow project conventions
• Reference past decisions
• Apply learned patterns

Manual Memory Management

// Save a memory entry
yield* Memory.save({
  key: "architecture-decision",
  content: "We use Event Sourcing for the order service",
  source: "user"
})

// List all memories
const memories = yield* Memory.list()

// Remove a memory
yield* Memory.remove({ key: "architecture-decision" })

Configuration

Memory settings in legion.json:

{
  "memory": {
    "enabled": true,
    "autoSave": true,
    "maxEntries": 100
  }
}

File Locations

File	Purpose
./LEGION.md	Project-specific memory
~/.legion/LEGION.md	Global memory
~/.legion/legion-memory.json	Session memory storage

Best Practices

1. Be specific - Write clear, actionable conventions
2. Update regularly - Keep LEGION.md current with project evolution
3. Use sections - Organize with clear headings
4. Include examples - Show code patterns when helpful
5. Date decisions - Track when decisions were made
6. Global vs project - Keep project-specific items in project LEGION.md

This site is open source. Improve this page.