memory-systems

Design cross-session agent memory with layered vector, graph, and temporal storage.

Updated Mar 13, 2026

Search Tags:#agents #retrieval #memory #persistence #graph #vector-store

Ready to run

npx skills add https://github.com/marinvch/ai-os --skill memory-systems-marinvch

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: memory-systems
Source: https://github.com/marinvch/ai-os/tree/main/.agents/skills/context-engineering-collection/skills/memory-systems
Command: npx skills add https://github.com/marinvch/ai-os --skill memory-systems-marinvch

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires numpy, and includes scripts (resource) and references (resource) components.

What problem does it solve?

Agents across sessions lose continuity; this skill provides a layered memory architecture that persists knowledge, links it to entities, and enables cross-session reasoning.

Core Features & Use Cases

Layered memory architecture combining vector stores for semantic search with graph-based entity relationships and temporal validity for time-sensitive facts.
Cross-session entity continuity, memory consolidation, and just-in-time memory loading to optimize context.
Real-world use cases include long-running conversations, multi-step assistant tasks, and cross-session knowledge graphs for users and entities.

Quick Start

Store a memory item for an entity and retrieve memories across sessions to verify cross-session continuity.

Frequently Asked Questions about memory-systems

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I implement cross-session memory for AI agents?▼

Cross-session memory for AI agents is implemented using a layered architecture that combines vector stores, property graphs, and temporal knowledge graphs to persist knowledge and enable continuity across diverse tasks.

What's the best way to store temporally-variant facts in a knowledge graph?▼

Storing temporally-variant facts is best handled by a temporal knowledge graph that tracks time-sensitive data validity, ensuring agents retrieve accurate entity relationships and memories across different sessions.

How does just-in-time memory loading work for long-running conversations?▼

Just-in-time memory loading optimizes context for long-running conversations by retrieving relevant entity memories from persistent storage only when needed, reducing overhead during multi-step assistant tasks.

Do I need numpy to build a vector store and graph memory architecture?▼

Yes, numpy is required as a dependency to support the numerical operations underlying the combined vector store, property graph, and temporal knowledge graph architecture for agent memory.

How do I consolidate duplicate entities across multi-session knowledge graphs?▼

Duplicate entities across multi-session knowledge graphs are resolved through entity deduplication and memory consolidation processes, linking stored items to maintain continuity for users and entities.