AI Memory SDK

A production-ready universal memory system for LLM applications.

Features

• Short-term memory: Sliding window message buffer.
• Long-term memory: Persistent storage for conversations.
• Semantic memory: Vector search using embeddings.
• Automatic summarization: Progressive conversation summaries.
• Multi-session: Isolated memory for different users.
• Token management: Smart context building within token limits.

Installation

npm install ai-memory-sdk

Quick Start

import { Memory } from 'ai-memory-sdk';

const memory = new Memory({
  sessionId: "user-123",
  vector: true,
  summarization: true
});

// Add messages
await memory.add("user", "My favorite color is green.");
await memory.add("assistant", "I will remember that!");

// Get context for LLM
const context = await memory.context();

// Search semantic memory
const history = await memory.search("What is my favorite color?");

Architecture

• core: Types and Context Builder.
• memory: Buffer, Vector, Summary, and Knowledge memory modules.
• storage: Pluggable storage (JSON, File, Redis).
• vector: Cosine similarity search.
• embeddings: OpenAI and Ollama providers.

License

MIT