Awesome Memory Systems

A curated list of awesome memory systems for AI agents and LLM applications.

Memory is what turns a stateless model into an agent that learns, remembers, and improves over time. This list collects libraries, frameworks, papers, and resources for giving AI systems short-term, long-term, and structured memory.

Origin: this list grew out of Ultimate Memory, writeit.ai's memory system for AI — most of the frameworks, papers, and benchmarks below served as research and inspiration for its design.

Contents

• Memory Frameworks & Libraries
• Building Blocks
• Papers & Research
• Evals & Benchmarks
• Related Lists

Memory Frameworks & Libraries

• Acontext — Agent skills as a memory layer; agents accumulate and reuse skills as persistent memory.
• AgentDB — Vector memory that gets smarter every time your agent uses it.
• Agentic Context Engine — Agents that learn from experience via evolving playbooks; open-source implementation based on the ACE (Agentic Context Engineering) paper.
• agentmemory — Persistent memory for AI coding agents (Claude Code, Copilot CLI, Cursor, Gemini CLI, Codex CLI and other MCP clients).
• Beads — A memory upgrade for your coding agent; graph-based issue tracking that doubles as durable agent memory.
• buildautomata_memory_mcp — Memory MCP server and CLI.
• claude-mem — Persistent context across sessions for every agent; captures everything your agent does.
• claude-supermemory — Lets Claude Code learn in real time, update its knowledge, and grow with you, backed by Supermemory.
• Cognee — Open-source AI memory platform for agents; turns raw data into knowledge graphs combined with vector search through modular ECL (extract–cognify–load) pipelines.
• company-brain — "GitHub for knowledge work": an open-source temporal knowledge graph and organizational memory engine.
• ConDB — The KV-cache-native context database, from the PageIndex team.
• cursor-memory-bank — Modular, documentation-driven persistent memory framework built on Cursor custom modes.
• DeepRefine — General LLM-based reasoning model for refining agent-compiled knowledge (HKUST KnowComp).
• Engram — DeepSeek's conditional memory via scalable lookup — a new axis of sparsity adding lookup-based memory inside LLMs (model-internal memory rather than an agent memory layer).
• engram — Persistent memory system for AI coding agents; agent-agnostic Go binary with SQLite + FTS5 and an MCP server (unrelated to DeepSeek's Engram above).
• gbrain — Garry Tan's opinionated OpenClaw/Hermes agent brain.
• Graphiti — Framework by Zep for building real-time, temporally-aware knowledge graphs as agent memory; tracks how facts change over time instead of only storing static snapshots.
• GraphRAG — Microsoft's modular graph-based RAG system; extracts an entity knowledge graph from unstructured text and uses community summaries for reasoning over large private corpora.
• HippoRAG — Neurobiologically inspired long-term memory framework for LLMs modeled on hippocampal indexing theory; improves multi-hop retrieval (associativity) and sense-making over complex contexts.
• HiRAG — Hierarchical graph-based RAG (EMNLP 2025 Findings); builds multi-level knowledge graphs by recursively clustering entity embeddings (HiIndex) and retrieves across local, bridge, and global levels (HiRetrieval).
• Hivemind — Turns agent traces into reusable skills shared across agents (Activeloop).
• HMLR — Living memory for AI.
• Honcho — Memory library for building stateful agents, centered on modeling users and maintaining theory-of-mind style representations across sessions.
• khive — A knowledge graph your AI agents build, query, and grow; for agents that need structure beyond vectors.
• Ladybug — Embedded property graph database built for query speed and scalability; a storage substrate for graph-based memory rather than a memory framework itself.
• Letta — Framework (formerly MemGPT) for building stateful agents with advanced memory that can learn and self-improve over time; includes an API, CLI, and agent runtime.
• LightMem — Lightweight and efficient memory-augmented generation (ICLR 2026).
• LightRAG — Simple and fast retrieval-augmented generation combining graph-based indexing with dual-level (local + global) retrieval.
• Mem0 — Universal memory layer for personalized AI; remembers user preferences and facts across sessions and adapts over time, with hosted and self-hosted options.
• Memobase — User-profile-based long-term memory for AI chatbot applications.
• MemoBrain — Executive memory for coherent long-horizon reasoning.
• memory-lancedb-pro — Enhanced LanceDB memory plugin for OpenClaw with hybrid retrieval (vector + BM25) and cross-encoder reranking.
• MemoryOS — Memory operating system for personalized AI agents, with OS-style hierarchical memory management (EMNLP 2025 Oral).
• MemPalace — Self-described "best-benchmarked open-source AI memory system"; free and open source.
• Memvid — Serverless, single-file memory layer for AI agents, positioned as a replacement for complex RAG pipelines.
• Momo — Self-hostable AI memory system written in Rust, inspired by Supermemory.
• OpenHuman — Personal AI that builds a local-first memory of your life.
• OpenKB — Open LLM knowledge base, from the PageIndex team.
• OpenMemory — Local persistent memory store for LLM applications, including Claude Desktop, GitHub Copilot, and Codex.
• OpenViking — The context database for AI agents (Volcengine/ByteDance).
• PageIndex — Vectorless, reasoning-based RAG; builds a hierarchical table-of-contents-like tree index of documents that LLMs navigate by reasoning instead of vector similarity.
• Recall — Durable project memory for Claude Code sessions, so you stop re-explaining your project every session.
• ReMe — "Remember Me, Refine Me": memory management kit for agents from the AgentScope ecosystem.
• Scout — Open-source company brain (Agno).
• Semantica — Accountability and context layer for AI: context graphs, decision intelligence, and full provenance.
• SimpleMem — Efficient lifelong memory for LLM agents, covering both text and multimodal memories.
• Supermemory — Memory and context engine behind a hosted API; the MIT-licensed repo contains the ecosystem around it — web app, browser/Raycast extensions, MCP server, and SDKs — while the engine itself is closed-source (Docker self-hosting offered separately).
• Sylph — Open-source company brain: run your company with AI agents, skills, and a self-improving context layer.
• TencentDB Agent Memory — Fully local long-term memory for AI agents built on a 4-tier progressive pipeline.

Building Blocks

Component libraries for building memory systems: chunking text, extracting facts, claims, entities, and relations from it, and resolving or deduplicating them.

Text Chunking

• semchunk — Fast, lightweight Python library for splitting text into semantically meaningful chunks.
• text-splitter — Rust and Python library for splitting text into semantic chunks up to a desired size, including token-based lengths.

Fact & Claim Extraction

• ClaimeAI — LangGraph-based fact-checking system; decomposes text into individual claims, verifies each against real-world evidence, and reports what's accurate.
• ClaimsMCP — The Claimify multi-stage claim-extraction methodology delivered as a local MCP server.
• claimify — Unofficial implementation of Microsoft's Claimify for high-quality claim extraction from LLM outputs.
• FActScore — Fine-grained atomic evaluation of factual precision in long-form text generation (EMNLP 2023).
• VeriScore — Pipelined factuality metric: claim extraction, evidence retrieval via web search, and claim verification.
• Molecular Facts — Code for "Molecular Facts: Desiderata for Decontextualization in LLM Fact Verification"; explores the right granularity for atomic facts.
• Long-Form Factuality (SAFE) — Google DeepMind's official code for "Long-form factuality in large language models", including the Search-Augmented Factuality Evaluator.

Entity & Relation Extraction

• GLiNER — Generalist and lightweight model for zero-shot named entity recognition.
• GLiREL — Generalist and lightweight model for zero-shot relation extraction; companion to GLiNER.

Knowledge Graph Construction

• DataFlow-KG — Knowledge graph data preparation with DataFlow-style operators and pipelines.
• Understand Anything — Turns any codebase, knowledge base, or docs into an interactive knowledge graph you can explore, search, and query.

Coreference Resolution

• fastcoref — Fast, accurate, and easy-to-use coreference resolution (F-COREF).
• Maverick — Efficient and accurate coreference resolution defying recent trends (ACL 2024).

Entity Resolution & Deduplication

• dedupe — Python library using machine learning and active learning for fuzzy matching, deduplication, and entity resolution on structured data.
• Splink — Fast, accurate, and scalable probabilistic record linkage and deduplication.
• Zingg — ML-based entity resolution for big data, built on Spark.

Papers & Research

Curated selection of papers on memory architectures for LLMs and agents. Dates are the first arXiv release (v1), ordered by release date, newest first.

• 2026-05-19 — Rethinking How to Remember: Beyond Atomic Facts in Lifelong LLM Agent Memory · code
• 2026-05-15 — DimMem: Dimensional Structuring for Efficient Long-Term Agent Memory · code
• 2026-05-13 — Useful Memories Become Faulty When Continuously Updated by LLMs
• 2026-05-13 — Cognifold: Always-On Proactive Memory via Cognitive Folding · code
• 2026-05-12 — SAGE: A Self-Evolving Agentic Graph-Memory Engine for Structure-Aware Associative Memory
• 2026-05-07 — From Storage to Experience: A Survey on the Evolution of LLM Agent Memory Mechanisms
• 2026-05-07 — STALE: Can LLM Agents Know When Their Memories Are No Longer Valid?
• 2026-04-17 — A Survey on the Security of Long-Term Memory in LLM Agents: Toward Mnemonic Sovereignty
• 2026-04-14 — GAM: Hierarchical Graph-based Agentic Memory for LLM Agents
• 2026-04-03 — Poison Once, Exploit Forever: Environment-Injected Memory Poisoning Attacks on Web Agents
• 2026-04-02 — Memory in the LLM Era: Modular Architectures and Strategies in a Unified Framework
• 2026-04-02 — DeltaMem: Towards Agentic Memory Management via Reinforcement Learning
• 2026-03-31 — MemFactory: Unified Inference & Training Framework for Agent Memory · code · dataset
• 2026-02-14 — Hippocampus: An Efficient and Scalable Memory Module for Agentic AI
• 2026-02-08 — Learning to Continually Learn via Meta-learning Agentic Memory Designs
• 2026-02-05 — Graph-based Agent Memory: Taxonomy, Techniques, and Applications · code
• 2026-02-02 — MemSkill: Learning and Evolving Memory Skills for Self-Evolving Agents · code · model
• 2026-02-02 — Beyond RAG for Agent Memory: Retrieval by Decoupling and Aggregation · code
• 2026-01-14 — Rethinking Memory Mechanisms of Foundation Agents in the Second Half: A Survey
• 2026-01-13 — AtomMem: Learnable Dynamic Agentic Memory with Atomic Memory Operation
• 2026-01-06 — MAGMA: A Multi-Graph based Agentic Memory Architecture for AI Agents
• 2026-01-06 — SYNAPSE: Empowering LLM Agents with Episodic-Semantic Memory via Spreading Activation
• 2026-01-05 — Agentic Memory: Learning Unified Long-Term and Short-Term Memory Management for Large Language Model Agents · code
• 2025-12-21 — MemEvolve: Meta-Evolution of Agent Memory Systems
• 2025-12-18 — MemoryGraft: Persistent Compromise of LLM Agents via Poisoned Experience Retrieval
• 2025-12-15 — Memory in the Age of AI Agents
• 2025-12-14 — Hindsight is 20/20: Building Agent Memory that Retains, Recalls, and Reflects
• 2025-11-23 — General Agentic Memory Via Deep Research · code
• 2025-10-06 — Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models
• 2025-09-29 — A-MemGuard: A Proactive Defense Framework for LLM-Based Agent Memory · code
• 2025-09-29 — MemGen: Weaving Generative Latent Memory for Self-Evolving Agents · code
• 2025-09-29 — ReasoningBank: Scaling Agent Self-Evolving with Reasoning Memory · code
• 2025-09-12 — HetaRAG: Hybrid Deep Retrieval-Augmented Generation across Heterogeneous Data Stores · code
• 2025-08-13 — Seeing, Listening, Remembering, and Reasoning: A Multimodal Agent with Long-Term Memory (M3-Agent) · code
• 2025-07-30 — Git Context Controller: Manage the Context of LLM-based Agents like Git
• 2025-07-10 — MIRIX: Multi-Agent Memory System for LLM-Based Agents · code
• 2025-06-09 — G-Memory: Tracing Hierarchical Memory for Multi-Agent Systems · code
• 2025-04-22 — From Human Memory to AI Memory: A Survey on Memory Mechanisms in the Era of LLMs
• 2025-03-13 — Retrieval-Augmented Generation with Hierarchical Knowledge (HiRAG) · code
• 2025-03-05 — Memory Injection Attacks on LLM Agents via Query-Only Interaction · code
• 2025-02-17 — A-MEM: Agentic Memory for LLM Agents · code
• 2025-02-17 — Unveiling Privacy Risks in LLM Agent Memory
• 2025-02-15 — Towards Effective Extraction and Evaluation of Factual Claims (Claimify)
• 2025-01-20 — Zep: A Temporal Knowledge Graph Architecture for Agent Memory · code
• 2024-12-12 — Memory Layers at Scale · code
• 2024-09-11 — Agent Workflow Memory (AWM) · code

Evals & Benchmarks

Benchmarks and evaluation frameworks for agent memory. Dates are the first arXiv release (v1), ordered by release date, newest first.

• 2026-05-18 — EvoMemBench: Benchmarking Agent Memory from a Self-Evolving Perspective · code
• 2026-05-12 — LongMemEval-V2: Evaluating Long-Term Agent Memory Toward Experienced Colleagues · code · dataset · leaderboard
• 2026-03-02 — AMemGym: Interactive Memory Benchmarking for Assistants in Long-Horizon Conversations · code · dataset
• 2026-02-22 — Anatomy of Agentic Memory: Taxonomy and Empirical Analysis of Evaluation and System Limitations · code
• 2026-02-18 — MemoryArena: Benchmarking Agent Memory in Interdependent Multi-Session Agentic Tasks · code · dataset
• 2026-02-11 — Locomo-Plus: Beyond-Factual Cognitive Memory Evaluation Framework for LLM Agents · code
• 2026-01-17 — MemoryRewardBench: Benchmarking Reward Models for Long-Term Memory Management in Large Language Models · code · dataset
• 2025-11-25 — Evo-Memory: Benchmarking LLM Agent Test-time Learning with Self-Evolving Memory
• 2025-10-27 — Evaluating Long-Term Memory for Long-Context Question Answering
• 2025-10-20 — MemoryBench: A Benchmark for Memory and Continual Learning in LLM Systems · code · dataset
• 2025-08-26 — Building Self-Evolving Agents via Experience-Driven Lifelong Learning: A Framework and Benchmark (StuLife) · code · dataset
• 2025-07-07 — Evaluating Memory in LLM Agents via Incremental Multi-Turn Interactions (MemoryAgentBench) · code · dataset
• 2025-02-18 — REALTALK: A 21-Day Real-World Dataset for Long-Term Conversation · code
• 2024-10-14 — LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory · code · dataset
• 2024-02-27 — Evaluating Very Long-Term Conversational Memory of LLM Agents (LoCoMo) · code

Related Lists

• awesome-ai-memory — A list of AI memory projects, maintained by the Cognee team.
• Agent-Memory-Paper-List — The paper list accompanying the survey "Memory in the Age of AI Agents".

Contributing

Contributions welcome! Please make sure the entry is actively maintained and add it in the format - [Name](link) — one-line description.