Why use a knowledge graph instead of a simple vector store for agent memory?

While vector stores are good for semantic similarity, they lose relationship structure. Knowledge graphs preserve links between entities, enabling agents to answer complex relational questions that vector RAG cannot.

What is a Temporal Knowledge Graph in the context of AI agents?

A Temporal Knowledge Graph adds validity periods (timestamps) to facts. This allows agents to understand when information was true, perform 'time-travel' queries, and prevent conflicts between old and new data.

How does this skill improve retrieval latency?

By using structured memory layers like Zep or GraphRAG, the system retrieves only relevant subgraphs or summaries rather than forcing the agent to process the entire conversation history, reducing latency by up to 90%.

What is the difference between working memory and short-term memory for agents?

Working memory is the immediate context window (volatile and zero-latency), while short-term memory typically persists across a single session via a database or cache, allowing search without overloading the context window.

Memory Systems

Name: Memory Systems
Author: muratcankoylan

bymuratcankoylan

•

7,140

•

数据科学与机器学习

Design and implement sophisticated agent memory architectures, from vector stores to temporal knowledge graphs, for cross-session persistence.

The Memory Systems skill provides a comprehensive framework for building AI agents that maintain continuity and reason over accumulated knowledge. It guides developers through the context-memory spectrum, covering working, short-term, and long-term storage layers. By implementing advanced patterns like Entity Tracking and Temporal Knowledge Graphs, this skill enables agents to preserve complex relationships, manage time-sensitive facts, and significantly reduce hallucinations. Whether you are optimizing retrieval latency with Zep or implementing GraphRAG for better reasoning, this skill provides the architectural blueprints for production-grade agent state management.

主要功能

01Memory consolidation strategies to manage unbounded growth and obsolete facts

02Implementation of Temporal Knowledge Graphs for time-aware querying

03Advanced retrieval patterns including GraphRAG and metadata-enhanced Vector RAG

04Multi-layer memory architecture design (Working, Short-term, Long-term, Permanent)

05Entity memory systems to maintain identity consistency across sessions

067,140 GitHub stars

使用场景

01Building persistent AI assistants that learn and adapt to user preferences over multiple sessions

02Developing complex multi-agent architectures that require shared, structured state and relationship tracking

03Creating enterprise knowledge bases that evolve over time while maintaining historical data integrity

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add muratcankoylan/agent-skills-for-context-engineering memory-systems

For use in Claude.ai and ChatGPT

Download Skill