memory-systems

Designs agent memory architectures and compares frameworks like Mem0, Zep, Letta, and Cognee.

Updated Mar 22, 2026
One-click install
npx skills add https://github.com/TECH-HY/SKILLS --skill memory-systems-tech-hy
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: memory-systems
Source: https://github.com/TECH-HY/SKILLS/tree/main/skills/memory-systems
Command: npx skills add https://github.com/TECH-HY/SKILLS --skill memory-systems-tech-hy

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Agents lose all state when sessions end, forcing users to repeat context and preventing long-term knowledge accumulation. This Skill guides the design of persistent memory architectures so agents retain entities, preferences, and facts across sessions. ## Core Features & Use Cases - Framework Comparison: Compares production memory frameworks (Mem0, Zep/Graphiti, Letta, LangMem, Cognee) with benchmark data from LoCoMo, LongMemEval, DMR, and HotPotQA. - Architecture Guidance: Provides decision tables for memory layers (working, short-term, long-term, entity, temporal knowledge graph) and retrieval strategies (semantic, entity-based, temporal, hybrid). - Use Case: When building a support agent that must remember a customer's history across conversations, use this Skill to choose between a simple file-system store and a temporal knowledge graph, then implement consolidation and error-recovery patterns. ## Quick Start Ask the agent to help you choose and implement a memory framework for an agent that must persist user preferences and entity facts across sessions.

Frequently Asked Questions about memory-systems

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

FAQPage Schema
How do I add long-term memory to an AI agent?

Start with file-system memory storing facts as structured JSON with timestamps, then move to Mem0 or a vector store when you need semantic search. Add Zep/Graphiti or Cognee only when you need relationship traversal or temporal validity tracking.

Mem0 vs Zep vs Cognee: which memory framework should I choose?

Mem0 offers the fastest path to production with managed infrastructure. Zep/Graphiti fits enterprise cases needing bi-temporal relationship modeling. Cognee builds denser multi-layer semantic graphs suited for multi-hop reasoning and evolving memory.

What is a temporal knowledge graph for agent memory?

A temporal knowledge graph stores facts with validity intervals (valid_from, valid_until), letting agents query what was true at a specific time. This prevents stale facts from poisoning context when information changes over time.

Does agent memory need a knowledge graph or is a vector store enough?

A vector store suffices for direct factual queries, but degrades on multi-hop reasoning. Benchmarks show Letta's simple filesystem agents scored 74% on LoCoMo, so add graph structure only when retrieval quality demands it.

Why does agent memory retrieval return stale or conflicting facts?

Stale results occur when temporal validity is not tracked. Check valid_until timestamps, prefer facts with the most recent valid_from, and trigger consolidation when most results are expired. Surface conflicts to users when confidence is low.

How do I benchmark an agent memory system?

Evaluate against LoCoMo, LongMemEval, or DMR benchmarks before and after changes. Treat results as signals for specific retrieval dimensions rather than definitive rankings, and monitor memory growth and retrieval latency in production.