Skip to content

memory-bank.md

Latest commit

 

History

History
58 lines (50 loc) · 1.99 KB

memory-bank.md

File metadata and controls

58 lines (50 loc) · 1.99 KB

Memory Bank Configuration

Overview

The Claude-Flow memory system provides persistent storage and intelligent retrieval of information across agent sessions. It uses a hybrid approach combining SQL databases with semantic search capabilities.

Storage Backends

  • Primary: JSON database (./memory/claude-flow-data.json)
  • Sessions: File-based storage in ./memory/sessions/
  • Cache: In-memory cache for frequently accessed data

Memory Organization

  • Namespaces: Logical groupings of related information
  • Sessions: Time-bound conversation contexts
  • Indexing: Automatic content indexing for fast retrieval
  • Replication: Optional distributed storage support

Commands

  • npx claude-flow memory query <search>: Search stored information
  • npx claude-flow memory stats: Show memory usage statistics
  • npx claude-flow memory export <file>: Export memory to file
  • npx claude-flow memory import <file>: Import memory from file

Configuration

Memory settings are configured in claude-flow.config.json:

{
  "memory": {
    "backend": "json",
    "path": "./memory/claude-flow-data.json",
    "cacheSize": 1000,
    "indexing": true,
    "namespaces": ["default", "agents", "tasks", "sessions"],
    "retentionPolicy": {
      "sessions": "30d",
      "tasks": "90d",
      "agents": "permanent"
    }
  }
}

Best Practices

  • Use descriptive namespaces for different data types
  • Regular memory exports for backup purposes
  • Monitor memory usage with stats command
  • Clean up old sessions periodically

Memory Types

  • Episodic: Conversation and interaction history
  • Semantic: Factual knowledge and relationships
  • Procedural: Task patterns and workflows
  • Meta: System configuration and preferences

Integration Notes

  • Memory is automatically synchronized across agents
  • Search supports both exact match and semantic similarity
  • Memory contents are private to your local instance
  • No data is sent to external services without explicit commands