agent-memory

Long-term memory layer for AI agents. Sessions give you multi-turn. This gives you multi-session.

Most agent memory libraries are vector search with no curation — they store everything, forget nothing, and silently contradict themselves. agent-memory treats memory as a managed asset with an explicit lifecycle: extraction, decay, forgetting, and contradiction resolution.

Installation

Packages are published under the @reaatech scope and can be installed individually:

# Main facade — batteries-included entry point
pnpm add @reaatech/agent-memory

# Individual packages for tree-shaking and direct use
pnpm add @reaatech/agent-memory-core
pnpm add @reaatech/agent-memory-storage
pnpm add @reaatech/agent-memory-embedding
pnpm add @reaatech/agent-memory-llm
pnpm add @reaatech/agent-memory-retrieval
pnpm add @reaatech/agent-memory-policies
pnpm add @reaatech/agent-memory-extraction
pnpm add @reaatech/agent-memory-events

Quick Start

import { AgentMemory, OpenAILLMProvider, MemoryType } from '@reaatech/agent-memory';

const memory = new AgentMemory({
  storage: { provider: 'memory' },
  embedding: {
    provider: 'openai',
    model: 'text-embedding-3-small',
    apiKey: process.env.OPENAI_API_KEY,
  },
  extraction: {
    llmProvider: new OpenAILLMProvider({
      apiKey: process.env.OPENAI_API_KEY,
      model: 'gpt-4o-mini',
    }),
    enabledTypes: [MemoryType.FACT, MemoryType.PREFERENCE, MemoryType.CORRECTION],
    batchSize: 10,
    confidenceThreshold: 0.7,
  },
});

// Extract and store memories from a conversation
const stored = await memory.extractAndStore(conversationTurns);

// Retrieve relevant memories for a query
const relevant = await memory.retrieve('Where does the user live?', { limit: 5 });

// Run maintenance — applies decay and forgetting policies
await memory.runMaintenance();

See the examples/ directory for working samples.

Packages

Package	Description
@reaatech/agent-memory	Main facade wiring all packages together
@reaatech/agent-memory-core	Core types, enums, and utilities
@reaatech/agent-memory-storage	Storage abstraction (In-Memory, PostgreSQL pgvector)
@reaatech/agent-memory-embedding	Embedding providers (OpenAI, Cohere, HuggingFace)
@reaatech/agent-memory-llm	LLM provider abstraction
@reaatech/agent-memory-retrieval	Semantic retrieval with ranking strategies
@reaatech/agent-memory-policies	Decay, forgetting, and contradiction resolution
@reaatech/agent-memory-extraction	LLM-based memory extraction from conversations
@reaatech/agent-memory-events	Event bus for memory lifecycle hooks

Features

• LLM-powered extraction — Identifies facts, preferences, decisions, and corrections from conversations with confidence scoring
• Semantic retrieval — Vector search with hybrid metadata filtering and five retrieval strategies (semantic, recency, importance, topic, adaptive)
• Lifecycle management — Exponential decay with configurable half-lives, automatic forgetting with capacity limits, and four contradiction resolution strategies
• Pluggable storage — In-memory adapter (dev/test) and PostgreSQL pgvector adapter (production)
• Multi-provider embedding — OpenAI, Cohere, and HuggingFace with transparent LRU caching
• Event hooks — Subscribe to memory lifecycle events for audit logging and metrics
• Multi-tenancy — Tenant-isolated memory spaces with per-tenant lifecycle management
• Custom policy rules — Define domain-specific retention rules (e.g., "medical preferences never decay")

Contributing

# Clone the repository
git clone https://github.com/reaatech/agent-memory.git
cd agent-memory

# Install dependencies
pnpm install

# Build all packages
pnpm build

# Run the test suite
pnpm test

# Lint and typecheck
pnpm lint
pnpm typecheck

Documentation

• ARCHITECTURE.md — System design, data models, and component interactions
• AGENTS.md — AI agent development guidelines and skills system
• CONTRIBUTING.md — Contribution workflow and release process
• SECURITY.md — Security policy

License

MIT