MemoryGate Documentation

Complete guide to building AI agents with persistent memory. MemoryGate is not storage—it's epistemic infrastructure. A living memory system that knows what it knows, why it knows it, and when it should let go.

System Highlights

  • Comprehensive MCP tools for observations, patterns, concepts, documents, and knowledge graphs
  • Semantic search with vector embeddings (pgvector + OpenAI text-embedding-3-small)
  • Hot/cold memory tiers with automatic lifecycle management and rehydration
  • Multi-tenant architecture with row-level security and shared memory stores
  • Polymorphic relationships linking any memory type to any other
  • Confidence-weighted memories with evidence chains and provenance tracking
  • OAuth 2.0 + PKCE authentication for secure MCP connections
  • Production-hardened: rate limits, audit logs, health checks, CSRF protection

Documentation Sections

Quick Start

Get up and running in 5 minutes. Connect your first AI agent and store your first memory.

MCP Setup

Configure MemoryGate with Claude Desktop, ChatGPT, or other MCP clients. Connection guides and troubleshooting.

Core Concepts

Understand the memory model: observations, patterns, concepts, and knowledge graphs. How retention and archiving work.

API Reference

Complete reference for all 41 MCP tools. Parameters, return types, and usage patterns for each endpoint.

Examples & Patterns

Common use cases and code examples. Learn from real-world implementations and best practices.

Getting Started

New to MemoryGate? Start with the Quick Start Guide to get your first agent connected. Already familiar with MCP? Jump straight to the API Reference.

Key Features

Semantic Memory: Your AI recalls meaning—not keywords. Intent, preference, history, nuance—retrieved by similarity, not strings. Memory that understands why, not just what.

Automatic Memory Lifecycle: High-signal memories stay hot. Low-confidence noise fades to cold storage—automatically. Signal survives. Clutter dies. History remains.

Living Knowledge Graphs: Observations, patterns, concepts, and documents link freely—across time and certainty. Memory that evolves instead of ossifies.

Confidence-Weighted Memory: Every memory carries certainty, provenance, and evidence chains. Your AI knows what it knows—and what it doesn't.

Append-Only Memory: Memories are never overwritten—only superseded with lineage intact. Reality leaves a paper trail.

Production-Hardened: OAuth 2.0, audit logs, rate limits, security headers—standard, enforced, boring. Because "experimental" is not a deployment strategy.

Architecture

MemoryGate is epistemic infrastructure—a memory system that knows what it knows, why it knows it, and when it should let go.

At its core, MemoryGate stores atomic observations with confidence scores, evidence chains, and domains. From these, it evolves understanding through patterns that synthesize reality as evidence accumulates, and concepts that remain canonical, alias-safe, and fragmentation-proof.

Memory chains preserve decision flow and causality. Polymorphic relationships connect observations, patterns, concepts, and documents—across time and certainty—building a living knowledge graph that breathes.

Hot/cold memory tiers handle lifecycle automatically: high-signal memories stay accessible, low-confidence noise archives gracefully, and rehydration brings context back on semantic demand.

Works alongside your agents through MCP—never inside them. No prompt injection, no model modification, no ToS violations. Your stack stays clean. Your agents get smarter.

Support & Community