npm create cortex-memories

persistent memory for AI agents

npm create cortex-memories

cortex-demo.app

Core Capabilities

Built for real-world AI systems

Persistent memory infrastructure for AI agents—from Memory Spaces and Hive Mode to ACID conversations and vector search. Everything your agents need to remember and coordinate.

Read the docs

4-Layer Architecture

ACID conversations, vector index, facts extraction, and graph database—fully interoperable with specialized APIs.

Memory Spaces

Flexible isolation per user, team, or project. Built-in support for multi-tenant AI applications.

User Profiles

Rich user preference tracking and context preservation across all agent interactions.

A2A Communication

Agent-to-agent messaging with importance scoring, delivery confirmation, and context propagation.

Context Chains

Hierarchical workflow coordination with graph-lite traversal for multi-agent orchestration.

Graph Database

Optional Neo4j/Memgraph integration for entity relationships and multi-hop knowledge traversal.

ACID conversations, vector index, facts extraction, and graph database—fully interoperable with specialized APIs.

Architecture

One API orchestrates the entire stack

cortex.memory.* automatically coordinates across ACID, Vector, Facts, and Graph. You call one method, all layers sync automatically.

Cortex architecture: cortex.memory.* orchestrating ACID, Vector, Facts, and Graph DB with automatic synchronization

Layer 1: ACID

Immutable Source

Append-only conversations. Never modified, kept forever. Perfect audit trail.

• Conversations (memorySpace)
• Immutable KB (shared)
• Mutable data (shared)

Layer 2: Vector

Searchable Index

Fast semantic search with embeddings. Links to ACID via conversationRef.

• Embeddings (any dimension)
• Semantic search
• Versioned (retention rules)

Layer 3: Facts

Extracted Knowledge

LLM-extracted structured facts. 60-90% storage savings, infinite context.

• Fact extraction
• Triple store (S-P-O)
• Graph sync ready

Layer 4: Unified API

Single Interface

cortex.memory.* orchestrates ALL layers automatically—ACID, Vector, Facts, Graph. One API, complete automation.

• Auto layer coordination
• Graph sync included
• Type-safe & tested

Explore the architecture

Features

Everything your AI needs to remember

From simple user preferences to complex multi-agent workflows. Cortex handles it all with enterprise-grade reliability.

Single API Layer

One unified cortex.memory.* interface for all operations. Simple, intuitive, and powerful.

Conversation Layer

Immutable ACID conversations with mutable sub-layer for preferences and profiles. Full history preservation.

Vector Index Layer

Fast semantic search with embedding support. Query millions of messages in milliseconds.

Facts Layer

LLM-powered extraction with 60-90% storage savings. Transform conversations into structured knowledge.

Graph Layer Integration

Ties relationships across all layers. Entity tracking, multi-hop traversal, and knowledge graphs.

GDPR Cascade Delete

One-click deletion across all layers with complete audit trail. Built-in compliance and governance.

Streaming Support

Native edge runtime compatibility. Works with Vercel AI SDK, OpenAI SDK, and LangChain.

MCP Server

Built-in Model Context Protocol server. Works with Cursor, Claude Desktop, and custom MCP clients.

Hive vs Collaboration

Hive Mode for shared memory (MCP, personal AI). Collaboration Mode for autonomous multi-agent systems.

One unified cortex.memory.* interface for all operations. Simple, intuitive, and powerful.

Use Cases

Built for real-world applications

From personal AI assistants to enterprise multi-agent systems, Cortex scales with your needs.

Chatbots & AI Assistants

Remember user preferences and conversation history across unlimited sessions.

User context preservation
Preference recall
Session continuity

Multi-Agent Systems

Coordinate between specialized agents with context chains and hive mode.

Agent coordination
Shared memory spaces
A2A communication

RAG Pipelines

Store and retrieve relevant context for LLM prompts with semantic search.

Semantic retrieval
Context injection
99% token reduction

Enterprise Support

Maintain customer context across interactions with GDPR compliance.

Customer history
Cascade deletion
Audit trails

Personal AI Tools

MCP integration for memory that follows you everywhere—Cursor, Claude, custom.

Cross-app memory
Zero duplication
MCP protocol

Knowledge Management

Organizational memory across teams with graph database integration.

Team workspaces
Graph queries
Knowledge graphs

Multi-Agent Architecture

Two modes. One powerful system.

Choose between shared memory spaces (Hive Mode) or isolated spaces (Collaboration Mode) based on your use case.

Hive Mode

Multiple agents share one memorySpace

Zero duplication—one memory serves all agents
Perfect for MCP cross-application memory
Instant consistency across all tools
Single write, everyone benefits

Perfect for:

Personal AI tools • Team workspaces • MCP integration

Collaboration Mode

Each agent has separate memorySpace

Complete isolation—prevents memory poisoning
Autonomous agents with independent memory
Secure cross-space access via Context Chains
A2A communication with audit trails

Perfect for:

Autonomous swarms • Enterprise workflows • Compliance

Infinite Context

Never run out of context again

Recall from millions of past messages via semantic search. Up to 99% token reduction compared to traditional context accumulation.

Token reduction visualization: Raw conversations (1.25M tokens) transformed to extracted facts (125K tokens) achieving 90% reduction for infinite context

Unlimited Recall

Access millions of memories from any point in history via semantic search

99% Savings

Token reduction through fact extraction means infinite context fits in finite windows

<100ms

Retrieve relevant memories from massive datasets with sub-second latency

memory.ts

import { Cortex } from '@cortexmemory/sdk'

// Initialize with Convex
const cortex = new Cortex({
  convexUrl: process.env.CONVEX_URL!
})

// Store with streaming (v0.9.0+)
const result = await cortex.memory.rememberStream({
  memorySpaceId: "user-123-personal",
  conversationId: "conv-1",
  userMessage: "What are best practices?",
  responseStream: stream, // Vercel AI SDK
  userId: "user-123",
  userName: "Alex",
  extractFacts: true // Auto fact extraction
})

// Search across millions of memories
const memories = await cortex.memory.search(
  "user-123-personal",
  "coding preferences",
  { enrichConversation: true }
)
// 99% token reduction via semantic retrieval

Developer Experience

Simple API.
Powerful architecture.

Built with developer experience in mind. Get started in minutes with npm create cortex-memories, scale to millions of memories with enterprise-grade reliability.