MemoryForge

MemoryForge provides a sophisticated memory management system for multi-agent AI applications. It enables agents to share knowledge, maintain context, and collaborate effectively.

3-Layer Memory Architecture

┌─────────────────────────────────────────────────────────────┐
│                    Agent Interface                           │
├─────────────────────────────────────────────────────────────┤
│  ┌─────────────────────────────────────────────────────┐   │
│  │ Layer 3: Episodic Memory (Short-term, Session)      │   │
│  │          SQLite - Fast access, temporary storage     │   │
│  └─────────────────────────────────────────────────────┘   │
│  ┌─────────────────────────────────────────────────────┐   │
│  │ Layer 2: Semantic Memory (Vector Embeddings)        │   │
│  │          Qdrant - Similarity search, dense vectors   │   │
│  └─────────────────────────────────────────────────────┘   │
│  ┌─────────────────────────────────────────────────────┐   │
│  │ Layer 1: Knowledge Graph (Persistent Relations)     │   │
│  │          Neo4j - Entity relationships, reasoning     │   │
│  └─────────────────────────────────────────────────────┘   │
└─────────────────────────────────────────────────────────────┘

Key Features

• Multi-Agent Support: Isolated and shared memory spaces
• REST API: Full CRUD operations for all memory types
• WebSocket: Real-time memory updates
• CLI Tools: Management and debugging utilities
• Backup/Restore: Data persistence and recovery
• Memory Compression: Automatic summarization of old memories

API Examples

# Store episodic memory
POST /api/v1/memory/episodic
{"agent_id": "agent-1", "content": "User prefers dark mode"}

# Query semantic memory
POST /api/v1/memory/semantic/search
{"query": "user preferences", "top_k": 5}

# Graph query
POST /api/v1/memory/knowledge/query
{"cypher": "MATCH (u:User)-[:PREFERS]->(p) RETURN p"}