https://www.youngju.dev/blog 천천히 올바르게. AI Researcher & DevOps Engineer Youngju's tech blog. GPU/CUDA, LLM, MLOps, Kubernetes AI workloads, distributed training, and data engineering. ko fjvbn2003@gmail.com (Youngju Kim) fjvbn2003@gmail.com (Youngju Kim) Sat, 16 May 2026 00:00:00 GMT https://www.youngju.dev/blog/culture/2026-05-16-ai-agent-memory-long-term-context-2026-mem0-zep-letta-cognee-graphiti-anthropic-memory-deep-dive.en https://www.youngju.dev/blog/culture/2026-05-16-ai-agent-memory-long-term-context-2026-mem0-zep-letta-cognee-graphiti-anthropic-memory-deep-dive.en Even with 1M-token context windows in 2026, agent memory is still an unsolved problem. This deep-dive compares Mem0, Zep, Letta (formerly MemGPT), Cognee, Anthropic Memory API, OpenAI Memory, Graphiti, Verba, Cody Memories, and the Generative Agents lineage. Three memory models — vector, graph, and episodic — are unpacked alongside short-term, working, and long-term (semantic + episodic + procedural) tiers. Storage backends from pgvector and Qdrant to Neo4j, Memgraph, and Kuzu are scored, plus the Korean (Upstage, NAVER HCX) and Japanese (Sakana, PFN) landscape — ending with explicit picks for chatbots, assistants, multi-agent systems, and research. Sat, 16 May 2026 00:00:00 GMT fjvbn2003@gmail.com (Youngju Kim) ai-memoryagent-memorymem0zeplettamemgptcogneegraphitianthropic-memoryopenai-memorygenerative-agentsverbavector-memoryknowledge-graph2026deep-diveenglish https://www.youngju.dev/blog/culture/2026-05-16-ai-agent-memory-long-term-context-2026-mem0-zep-letta-cognee-graphiti-anthropic-memory-deep-dive.ja https://www.youngju.dev/blog/culture/2026-05-16-ai-agent-memory-long-term-context-2026-mem0-zep-letta-cognee-graphiti-anthropic-memory-deep-dive.ja コンテキストウィンドウが100万トークンを超えた2026年でも、エージェントメモリは未解決の問題だ。Mem0、Zep、Letta（旧MemGPT）、Cognee、Anthropic Memory API、OpenAI Memory、Graphiti、Verba、Cody Memories、そしてGenerative Agentsまで — ベクトルメモリ・グラフメモリ・エピソードメモリの3モデルを比較し、短期・作業・長期（意味・エピソード・手続き的）メモリの階層を解剖する。pgvector / Qdrant / Neo4j / Memgraph / Kuzuといったストレージバックエンド、韓国（Upstage、NAVER HCX）と日本（Sakana、PFN）の動向まで含め、チャットボット・アシスタント・マルチエージェント・研究のシナリオ別にどのメモリシステムを選ぶべきかを決める。 Sat, 16 May 2026 00:00:00 GMT fjvbn2003@gmail.com (Youngju Kim) ai-memoryagent-memorymem0zeplettamemgptcogneegraphitianthropic-memoryopenai-memorygenerative-agentsverbavector-memoryknowledge-graph2026deep-dive日本語 https://www.youngju.dev/blog/culture/2026-05-16-ai-agent-memory-long-term-context-2026-mem0-zep-letta-cognee-graphiti-anthropic-memory-deep-dive https://www.youngju.dev/blog/culture/2026-05-16-ai-agent-memory-long-term-context-2026-mem0-zep-letta-cognee-graphiti-anthropic-memory-deep-dive 컨텍스트 윈도우가 100만 토큰을 넘긴 2026년에도 에이전트 메모리는 여전히 풀리지 않은 문제다. Mem0, Zep, Letta(구 MemGPT), Cognee, Anthropic Memory API, OpenAI Memory, Graphiti, Verba, Cody Memories, Generative Agents까지 — 벡터 메모리·그래프 메모리·에피소딕 메모리 세 가지 모델을 비교하고, 단기·작업·장기(시맨틱·에피소딕·절차적) 메모리 계층을 풀어낸다. pgvector / Qdrant / Neo4j / Memgraph / Kuzu 저장 백엔드, 한국(Upstage·NAVER HCX)과 일본(Sakana·PFN)의 흐름까지, 챗봇·어시스턴트·멀티에이전트·연구 시나리오별로 어떤 메모리 시스템을 골라야 하는지 정한다. Sat, 16 May 2026 00:00:00 GMT fjvbn2003@gmail.com (Youngju Kim) ai-memoryagent-memorymem0zeplettamemgptcogneegraphitianthropic-memoryopenai-memorygenerative-agentsverbavector-memoryknowledge-graph2026deep-dive