@rdk-moss/memory

Context-aware memory management for the Moss agent runtime.

Features

• BM25 + Embedding Hybrid Search: Combines keyword-based BM25 scoring with semantic embedding similarity for robust memory retrieval
• Scope-based Organization: Memories organized by workspace, user, device, and learning scopes
• Self-learning Memory: Automatic extraction and storage of user preferences and corrections
• Cross-language Recall: Query in one language, retrieve memories stored in another
• Staleness Management: Automatic detection and decay of stale memories

Installation

npm install @rdk-moss/memory

Usage

import { MemoryManager } from '@rdk-moss/memory';

const memory = new MemoryManager('./.dmoss/memory', optionalEmbeddingProvider);

// Store a memory
await memory.add('User prefers TypeScript over JavaScript', {
  scope: 'user',
  pinned: true,
});

// Search memories
const results = await memory.search('programming language preference', {
  scope: 'user',
  limit: 5,
});

// Load context for agent
await memory.loadContext();

API

• MemoryManager: Main memory storage and retrieval
• MemoryContextSelector: Context-aware memory selection for agent turns
• SelfLearningMemory: Automatic preference and correction extraction
• WorkspaceMemory: Workspace-level memory management

License

MIT