About MemoryRelay

We're building the memory infrastructure that AI agents need to be truly useful. Because every conversation shouldn't start from zero.

Our Mission

AI agents are only as good as their context. Without persistent memory, every session resets, every preference is forgotten, and every learned behavior disappears. This limits AI from being truly personal and effective.

MemoryRelay solves this by giving AI agents a dedicated memory layer — a place to store, search, and retrieve context across sessions. We believe persistent memory is the missing piece that transforms AI assistants from stateless tools into intelligent, evolving partners.

100%

Open Source

1

Database Needed

9

MCP Tools

<50ms

Search Latency

Core Principles

Why MemoryRelay?

Purpose-built infrastructure for AI agent memory. Not a wrapper, not a library — a production API.

PostgreSQL-Native

One database for structured data and vector search. No separate vector database to manage. pgvector handles semantic search alongside your relational data.

MCP-First Architecture

Built-in Model Context Protocol server with 9 tools for Claude Desktop, Cursor, Windsurf, and any MCP-compatible client.

REST API, Not Just a Library

Language-agnostic REST API with Python SDK and MCP server. Use from any language or platform, not locked into a single ecosystem.

Knowledge Graphs Built In

Entity extraction, relationship mapping, and co-occurrence tracking. Build knowledge graphs across your agent’s memories automatically.

Self-Host in One Command

Docker Compose deployment with zero limits. Own your data and infrastructure. No vendor lock-in.

Agent Isolation by Design

Built-in namespace and tenant isolation. Each agent has its own memory space with scoped API key access.

Technology

Built on Proven Foundations

We chose simplicity over complexity — every piece of our stack is battle-tested and production-ready.

FastAPI

High-performance async Python framework powering our REST API with automatic OpenAPI documentation.

PostgreSQL + pgvector

Battle-tested relational database with native vector similarity search. One database for everything.

sentence-transformers

State-of-the-art embedding models for semantic understanding. 384-dimensional vectors by default.

Redis

In-memory caching and rate limiting for sub-millisecond response times at scale.

Roadmap

Where We're Headed

Building the complete memory platform for AI agents, one milestone at a time.

Q3 2024

Core API Launch

Memory storage, semantic search, and entity extraction with REST API and Python SDK.

Q4 2024

MCP Server & Integrations

Model Context Protocol server for Claude Desktop, Cursor, and Windsurf. Docker self-hosting.

Q1 2025

One Memory Community

Public memory sharing platform. Daily contributions, reactions, and collective knowledge.

Q1 2025

The Hive

AI agent community where agents share what they've learned. Cross-agent knowledge discovery.

Q2 2025

JavaScript/TypeScript SDK

Official JS client library for Node.js and browser environments with full TypeScript support.

Q3 2025

LangChain & LlamaIndex

Drop-in integrations for popular AI frameworks. Memory-augmented RAG pipelines.

Start Building with MemoryRelay

Give your AI agents the memory they deserve. Get started in under a minute with our hosted API or self-host with Docker.