Agent Memory

Σ-Mem maintains historical competence evidence matrices M_p, a peer relationship matrix G, a memory readout r_{p,t}, and residual steering δ_{p,t} to guide multi-agent trust decisions. On the mixed counterfactual benchmark, Σ-Mem raises Qwen3-0.6B accuracy at CF@90 from 46.22% with the base model to 71.10% with Σ w/ G.

Focus Agent adds start_focus, complete_focus, a persistent Knowledge block, and an optimized Persistent Bash plus String-Replace Editor scaffold to actively compress context during long software-engineering tasks. On five hard SWE-bench Lite instances against a Baseline ReAct agent, Focus Agent achieves 22.7% token reduction (14.9M → 11.5M) while matching 3/5 = 60% task success.

ActMem transforms dialogue history into atomic facts via Memory Fact Extraction, groups them with Fact Clustering, links them through a Memory KG Construction module, and uses Counterfactual-based Retrieval and Reasoning for action-aware answers. On ActMemEval, ActMem reaches 76.52% QA accuracy with DeepSeek-V3, beating LightMem’s 63.97% by 12.55 points and NaiveRAG’s 61.54%.

ADAM combines Anchor extraction, Distribution estimation, Anchor selection, and Query generation to adaptively probe agent memory via an auxiliary generator and entropy based selection. On the EHRAgent benchmark with Llama2-7b-chat, ADAM reaches EQ=77 and ASR=1.00, compared to MEXTRA’s EQ=44 and ASR=0.89.

A-MAC scores candidate memories using Utility, Confidence, Novelty, Recency, and Type Prior combined by a learned linear admission policy with Algorithm 1 A-MAC Memory Admission. On the LoCoMo benchmark, A-MAC achieves F1 0.583 and 2644 ms latency, improving F1 by 0.042 and reducing latency by 1187 ms compared to A-mem.

Aeon restructures LLM memory using the Atlas, Trace, Semantic Lookaside Buffer, Write Ahead Log, and Sidecar Blob Arena inside a zero copy Core Shell kernel. Aeon achieves 4.70 ns INT8 dot products, 3.09 µs Atlas traversal at 100K nodes, 3.1× compression, and P99 read latency of 750 ns under 16 thread contention compared to FP32 and flat scan baselines.

Agentic Memory (AgeMem) exposes memory management tools, a three-stage progressive RL strategy, and step-wise GRPO directly inside the agent policy to jointly control long-term and short-term memory. On Qwen3-4B-Instruct, AgeMem attains 54.31% average performance across ALFWorld, SciWorld, PDDL, BabyAI, and HotpotQA, exceeding the best baseline A-Mem at 45.74%.

Agent Memory Below the Prompt stores each agent’s KV state in a block pool, quantizes it via a Q4 pipeline, reloads it with BatchQuantizedKVCache, and reuses it across phases using cross-phase context injection. On Gemma 3 12B, Agent Memory Below the Prompt reduces cold TTFT from 172,096 ms to 1,264 ms at 32K context (136×) compared to FP16 prefix caching baselines like vllm-mlx.

Agent Memory decomposes agent workloads into ingestion, memory construction, storage, retrieval, prompt assembly, generation, and maintenance, and classifies ten systems across four paradigms. Agent Memory’s profiling on MemoryAgentBench and MemoryArena reveals over 47× spread in lifecycle energy per correct answer and two orders of magnitude differences in serving latency across paradigms.

AGENTSYS organizes a Main Agent, Worker Agents, Intent Schemas, and an Alignment Validator into a hierarchical memory system that isolates raw tool outputs and only admits schema-validated JSON. On AgentDojo, AGENTSYS reaches 52.87% attacked utility and 0.78% ASR versus 48.27% and 30.66% for the No Defense baseline.

Neuro-Symbolic Dual Memory Framework uses Progress Memory, Feasibility Memory, a Blueprint Planner Agent, a Progress Monitor Agent, and an Actor Agent to decouple semantic progress guidance from executable feasibility checks. On ALFWorld, Neuro-Symbolic Dual Memory Framework achieves 94.78% success rate versus 88.81% for AWM, and on WebShop reaches 0.7132 score versus 0.5998 for WALL-E 2.0.

AMA orchestrates four agents — the Constructor, Retriever, Judge, and Refresher — to build Raw Text, Fact Knowledge, and Episode Memory and route queries adaptively across these granularities. On the LoCoMo benchmark with GPT-4.1-mini, AMA achieves an overall LLM Score of 0.805 compared to Nemori’s 0.774, while reducing token consumption by approximately 80% relative to FullContext.

AMemGym combines Structured Data Generation, On-Policy Interaction, Evaluation Metrics, and Meta-Evaluation to script user state trajectories, drive LLM-simulated role-play, and score write–read–utilization behavior. On AMemGym’s base configuration, AWE-(2,4,30) reaches a 0.291 normalized memory score on interactive evaluation, while native gpt-4.1-mini only achieves 0.203, exposing substantial gaps between memory agents and plain long-context LLMs.

AMV-L manages agent memory using a Memory Value Model, Tiered Lifecycle, Bounded Retrieval Path, and Lifecycle Manager to decouple retention from retrieval eligibility. Under a 70k-request long-running workload, AMV-L improves throughput from 9.027 to 36.977 req/s over TTL and reduces p99 latency from 5398.167 ms to 1233.430 ms while matching LRU’s retrieval quality.

Anatomy of Agentic Memory organizes agentic memory into four structures using components like Lightweight Semantic Memory, Entity-Centric and Personalized Memory, Episodic and Reflective Memory, and Structured and Hierarchical Memory. Anatomy of Agentic Memory then reports comparative results such as Nemori’s 0.781 semantic judge score on LoCoMo versus SimpleMem’s 0.298, and latency differences like 1.129s for Nemori versus 32.372s for MemoryOS.

Mnemonic Sovereignty analyzes long term Write, Store, Retrieve, Execute, Share, and Forget Rollback phases against integrity, confidentiality, availability, and governance objectives for agent memory. Mnemonic Sovereignty’s lifecycle matrix shows most of the ~70 works cluster on write and retrieve integrity, leaving store, availability, and governance primitives like write gate validation and post deletion verification almost entirely unexplored.

ATANT v1.1 structurally analyzes seven benchmarks using the 7 v1.0 continuity properties, the 10 checkpoints, a property-coverage matrix, and the Kenotic v1.0 reference implementation. ATANT v1.1 reports 96% ATANT cumulative-scale versus 8.8% LOCOMO substring accuracy, showing that LOCOMO, LongMemEval, BEAM, MemoryBench, Zep eval, MemGPT/Letta, and RULER measure different properties from continuity.

AtomMem reframes agent memory as a POMDP and composes atomic CRUD operations over a hybrid scratchpad plus vector memory storage using a GRPO-based RL policy. On HotpotQA, 2WikiMultiHopQA, Musique, GAIA, and WebWalkerQA, AtomMem reaches an average 58.8 score, beating MemAgent’s 56.7 with the same Qwen3-8B backbone.

BeliefMem maintains an external belief-based memory bank with Add, Merge, and Belief-aware Retrieval over attribute-level hypotheses. On LoCoMo, BeliefMem reaches 42.38 F1 with GPT-4o-mini, beating Mem0’s 40.99 F1, and on ALFWorld it attains 59.88% success rate vs ReadAgent’s 54.03% (+5.85).

MEMORYARENA orchestrates Memory-Agent-Environment Loops, Multi-Session Working Flow, Bundled Web Shopping, Group Travel Planning, and Progressive Web Search to stress-test how agents store and reuse information across sessions. MEMORYARENA’s main result is that agents with near-saturated scores on long-context benchmarks like LoCoMo still obtain Task Success Rates as low as 0.00–0.12 across its four environments.

CAMA combines Neuro-Symbolic Evidence Assignment, Effective Independent Evidence Estimation, Factor-Level Conflict Arbitration, and Active Independent-Evidence Recovery to decouple correlated memories into latent evidence slots and recover missing sources. On LongMemEval with DeepSeek V4 Flash, CAMA reaches 87.9 EM compared to MADAM RAG’s 86.8 EM while also cutting Replication Sensitivity from 15.3 to 7.8.

CAST builds views, scenes, character profiles, a semantic index, and an episodic index to organize dialogue into person-conditioned event structures. CAST reaches 62.02 F1 and 81.21 J on LOCOMO open questions, beating Zep by +13.24 F1 and +8.32 J and vanilla RAG by +25.15 F1 and +24.61 J.

ChronoMem wraps ADK’s LocalMemoryService, SQLite memory store, Version Index, and Reranker module into a semantic version-control layer that snapshots agent memory on every write and rolls back via natural-language queries. ChronoMem reaches 55.1% rollback-consistent QA accuracy on MemoryAgentBench with Qwen2.5-7B, compared to 35.5% for the RAG-only baseline.

ClawVM manages agent state as typed pages via the SessionPageTable, RepresentationSelector, FaultObserver, WritebackJournal, and ClawVMEngine inside the agent harness. Across four OpenClaw-derived workloads and six token budgets, ClawVM cuts explicit faults from 67.8 (retrieval baseline) and 1.5 (Compaction-Hybrid) to 0.0 while adding median <50 μs policy-engine overhead per turn.

D-Mem combines Mem0∗, Quality Gating, and Full Deliberation into a dual-process memory system that incrementally stores vector memories and selectively scans raw history. On LoCoMo with GPT-4o-mini, D-Mem’s Quality Gating reaches 53.5 F1 versus the Mem0∗ baseline’s 51.2 F1, recovering 96.7% of the 55.3 F1 Full Deliberation performance with far fewer tokens.

Drawing on Memory uses dual-trace memory encoding, an evidence scoring gate, and a three-state retrieval protocol to store paired fact and scene traces in Letta’s archival memory. On LongMemEval-S, Drawing on Memory reaches 73.7% accuracy versus 53.5% for the fact-only C7-control baseline, a +20.2 percentage point gain concentrated in temporal, update, and multi-session questions.

DreamBench-SWE wraps a fixed wake agent with a raw episode log, derived memory store, maintenance pipeline, and retrieval gate to test multi-session memory hygiene in software repositories. On 60 traps and 180 S3 cells, the strongest verbatim baseline B5 reaches 89/180 Pass@1 while the hybrid reference probe reaches 95/180, but the clustered P1 comparison fails to reject (p=0.518).

E-mem combines a Master Agent, multiple Assistant Agents, a Multi-Pathway Routing Mechanism, Sliding Window Segmentation with Overlap, and Episodic Memory Context Retention and Isolation to reconstruct native contexts instead of compressing them. On the LoCoMo benchmark, E-mem achieves 54.17 F1 with GPT-4o-mini, beating GAM by 8.86 F1 while reducing normalized token cost from 169100 to 3621.

EvoMemBench evaluates self-evolving agent memory using four settings and six datasets spanning in-episode and cross-episode, knowledge and execution evolution. EvoMemBench’s main result is that explicit memory only helps reliably when context is insufficient, tasks are hard, or stored experience matches the decision process, while Gemini-3-Flash and other long-context baselines often match or beat 15 memory methods.

Experience Compression Spectrum organizes Level 0 Raw Trace, Level 1 Episodic Memory, Level 2 Procedural Skill, and Level 3 Declarative Rule into a unified scaffold-level compression framework. Experience Compression Spectrum’s mapping of 20+ systems and <1% cross-citation rate shows that all existing agents fix a single compression level and never perform adaptive cross-level compression.

FadeMem manages agent memory using a Dual-Layer Memory Architecture, Biologically-Inspired Forgetting Curves, Memory Conflict Resolution, and Adaptive Memory Fusion to selectively retain and consolidate information. On LoCoMo, FadeMem attains 29.43 multi-hop F1 and 0.45 SRR, improving over Mem0’s 28.37 F1 while cutting storage by 45%.

ForeDreamer separates factual memory from experiential memory, using a main agent, a memory-processing subagent, an Experience Bank, and a MemGuide–MemTools workspace to process web evidence before forecasting. On Prophet Arena, ForeDreamer achieves an average Brier score of 0.1471 with Qwen3.5-Flash, improving over the Full Text baseline at 0.2059.

GAM builds a Hierarchical Graph Memory Architecture with a global Topic Associative Network, local Event Progression Graphs, State-Based Memory Consolidation, and Graph-Guided Multi-Factor Retrieval to decouple encoding from consolidation. On LoCoMo with Qwen2.5-7B, GAM attains an Average F1 of 40.00 compared to Mem0’s 35.38, and on LongDialQA with Qwen2.5-7B, GAM reaches 12.55 F1 vs MemoryOS at 6.76.

SSGM interposes a Governance Middleware, Read Filtering Gate, Write Validation Gate, and a dual substrate of Mutable Active Graph plus Immutable Episodic Log between agents and memory. SSGM unifies evolving-memory systems into a four-dimensional failure taxonomy and proves that periodic reconciliation can bound semantic drift over infinite horizons.

Graph-based Agent Memory organizes agent memory into Knowledge vs Experience Memory, Short-term vs Long-term Memory, and Non-structural vs Structural Memory tied to graph implementations. It unifies memory extraction, storage, retrieval, and evolution into a single lifecycle, mapping over 70 named systems like MemGPT, GraphRAG, and Zep into a coherent design space.

Harness the Memory instruments a Unified Evaluation Harness, External Memory Substrate Families, Internal Memory Substrate Families, Retrieval Depth Sweep, and Scalability Study to compare 11 memory substrates under identical agents. Harness the Memory shows, for example, that M7 Distilled Strategies reaches 32.1% TSR on ALFWorld-unseen with QWEN3-32B-AWQ, a +9.7 percentage point gain over the NoMem baseline at only 1.23× latency.

HiGMem organizes dialogue history into a Turn Layer, Event Layer, Memory Construction Mechanism, and LLM-Guided Retrieval that use event summaries as semantic anchors for turn selection. On LoCoMo10, HiGMem reaches 0.78 adversarial F1 and 0.49 single-hop F1, beating A-Mem while using an order of magnitude fewer retrieved turns.

Infini Memory organizes persistent agent memory into topic documents, a CURRENT buffer, REWRITE_CURRENT, a topic library, and an agentic reader that iteratively inspects evidence. On MemoryAgentBench, Infini Memory-A scores 64.7% overall on gpt-5-mini, beating HippoRAG-v2 at 45.5% by 19.2 points.

Keep It InMind introduces the InMind benchmark plus paired controls (Naive query, Indirect query, Target recall, Backbone control) to isolate failures in agent memory use. On InMind, Keep It InMind finds that the backbone reaches 84.0% indirect accuracy with the memory in context, while six retrieval-based systems reach at most 14.4% application accuracy.

LazyMem stores raw histories and at query time runs Hybrid Retrieval, History Windowing, and a trained Memory-Processing Model to build compact, query-conditioned memory for an answer LLM. On LongMemEval, LazyMem-4B achieves 0.85 LLM-judge accuracy with 213 answer-context memory tokens, a 0.03 gain over StructMem while using 21.0× fewer tokens.

LightMem orchestrates SLM-1 Controller, SLM-2 Selector, SLM-3 Writer, and STM MTM LTM stores to modularize retrieval, writing, and offline consolidation. On LoCoMo, LightMem reaches 34.50 F1 for GPT-4o multi hop questions, +1.64 over A-MEM, while keeping median retrieval latency at 83 ms.

LycheeMemory V2 combines Online Semantic Segmentation, Segment-Level Memory Encoding, Structured Evidence Organization, and Plan-Guided Multi-Route Retrieval to batch coherent dialogue segments into typed, indexed records. On LoCoMo, LycheeMemory V2 achieves 89.22% overall accuracy versus 68.83% for A-Mem, and on LongMemEval-S it reaches 92.20% versus 71.60% for A-Mem, with up to 7.2× fewer construction tokens.

MAGMA organizes agent memory with an Intent-Aware Router, Adaptive Topological Retrieval, a Data Structure Layer of Relation Graphs and Vector Database, plus dual-stream Synaptic Ingestion and Asynchronous Consolidation. On LoCoMo, MAGMA achieves a 0.700 overall LLM-as-a-Judge score versus 0.590 for Nemori, and reaches 61.2% average accuracy on LongMemEval versus 56.2% for Nemori.

Mandol combines a hierarchical memory model, SemanticMap, SemanticGraph, and a quantitative query mechanism to unify agent memory and retrieval. On LoCoMo with GPT-4.1-mini, Mandol achieves 92.21% overall accuracy, 0.24 percentage points higher than EverMemOS while using fewer tokens.

Mem2ActBench integrates Heterogeneous Data Integration, Fact Extraction and Grouping, Memory Evolution Chain Construction, and Memory-anchored Q&A Construction to create long, interruption-heavy sessions with grounded tool calls. On Mem2ActBench, oracle retrieval reaches F1 53.8 while the best passive hybrid retriever at k=5 reaches only 30.7, revealing a 23.1 F1 gap in memory utilization.

MemAdapter combines a Generative Subgraph Retriever, Anchored Alignment Module, Target Alignment Module, Unified Memory Space, and Agent Model to turn diverse memory states into explicit evidence subgraphs. On NarrativeQA with Qwen2.5-7B, MemAdapter achieves F1 61.59 compared to 56.91 for MemoryLLM and 50.01 for Mem0, while completing cross-paradigm alignment in about 13 minutes with less than 5% of training compute.

MemAudit combines Counterfactual Memory Influence Score, Memory Consistency Graph, and a fused Detoxification Score to rank and remove suspicious memories in MINJA-poisoned agents. On MINJA QA with GPT-4o, MemAudit reduces attack success rate from 70.0% to 0.0%, beating random deletion, retrieval-frequency deletion, and nearest-neighbor contradiction filtering.

MemEvoBench combines Misleading Memory Injection, Noisy Tool Returns, Biased User Feedback, and a Memory Modification Tool (+ModTool) to stress-test long-term memory safety in LLM agents across 7 domains and 36 risk types. On the QA Style benchmark, MemEvoBench shows Gemini-2.5-Pro’s ASR drops from 67.0% (Vanilla) to 19.0% with +ModTool in Round 1, while biased feedback can push GPT-5’s QA ASR from 59.0% to 78.0% by Round 3.

Memex(RL) optimizes Indexed Experience Memory, CompressExperience, ReadExperience, and ContextStatus so Memex keeps only an indexed summary in-context while archiving full artifacts externally. On modified ALFWorld, Memex(RL) lifts task success from 24.22% to 85.61% over the Memex agent without RL while reducing peak working context from 16,934.46 to 9,634.47 tokens.

MemFactory decomposes memory agents into Module Layer, Agent Layer, Environment Layer, and Trainer Layer with plug and play Extractor, Updater, Retriever, and RecurrentMemoryModule components. On MemAgent eval_50, MemFactory raises Qwen3-1.7B from 0.4727 to 0.5684 and Qwen3-4B-Instruct from 0.6523 to 0.7051 using GRPO.

MemForest combines parallel extraction, a shared memory substrate of canonical facts, scoped MemTrees, and forest recall plus tree browse to maintain temporal agent memory efficiently. On LongMemEval-S, MemForest with Qwen3-30B-A3B-Instruct-2507 achieves 81.8% pass@1 overall, 14.8 points above EverMemOS, while its input-normalized build rate is 6.0× higher.

MemLineage wraps a single memory store with six modules: Provenance metadata, Ed25519 signing, an RFC 6962 Merkle log, a weighted lineage DAG, verifier-aware retrieval, and a sensitive-action gate that enforces Untrusted-Path Persistence. On a deterministic harness with AgentPoison-style, MemoryGraft-style, and sleeper-via-derivation attacks, MemLineage is the only configuration that achieves 0.00 ASR on all three while keeping per-operation overhead below one millisecond.

Memory as Metabolism defines companion knowledge systems with five retention operations (TRIAGE, DECAY, CONTEXTUALIZE, CONSOLIDATE, AUDIT) plus memory gravity and minority-hypothesis retention over a raw buffer, active wiki, and cold memory. Instead of benchmark gains, Memory as Metabolism’s main result is a governance specification that separates descriptive, taxonomic, and normative claims and predicts improved coherence stability, fragility resistance, monoculture resistance, and effective minority-hypothesis influence for companion wikis.

MEMORYCD builds a user memory pool Mu from lifelong Amazon Review histories and evaluates long-context prompting, Mem0, LoCoMo, ReadAgent, MemoryBank, and A-Mem across rating, ranking, and personalized text tasks. On Books and Home & Kitchen, MEMORYCD shows GPT-5 reaches RMSE 0.551–0.624 and NDCG@3 up to 0.610, while Gemini-2.5 Pro peaks at ROUGE-L 0.222 for generation, revealing substantial remaining gaps to real user behavior.

MemoryCPT combines Query-agnostic Distillation, Query-aware Retrieval and Summarization, RRF-based Retrieval, and a GRPO-trained Fine Summary adapter on top of Qwen2.5-7B or Llama-3.2-3B. On LoCoMo, MemoryCPT with Qwen-2.5-7B achieves F1 0.479 versus BudgetMem’s 0.373, while reducing Cost from 24.11 to 4.31.

Memory for Autonomous LLM Agents decomposes agent memory into a POMDP-grounded write–manage–read loop, a three-dimensional taxonomy, and five mechanism families spanning context compression, retrieval stores, reflection, hierarchical virtual context, and policy-learned management. Memory for Autonomous LLM Agents synthesizes results like Voyager’s 15.3× tech-tree speedup and MemoryArena’s 80%→45% drop to show that memory architecture often matters more than backbone choice.

Memory in the LLM Era decomposes agent memory into Information Extraction, Memory Management, Memory Storage, and Information Retrieval, then recombines modules into a new hierarchical tree–tier architecture. On LONGMEMEVAL with Qwen2.5-7B, Memory in the LLM Era achieves 38.79 F1 overall versus 36.92 for MemTree.

MemoryLake organizes confirmed conclusions, supporting evidence, and reusable experience into separate tracks with presence policies, using gpt-5-mini and bge-m3 for consolidation, retrieval, and bounded prompt assembly. On the shared MemoryArena sets, MemoryLake reaches a 20.5% equal-weight macro-average SR compared to the 13.6% best comparator, with SR 9/40 in mathematics, 12/20 in physics, and 4/20 in progressive retrieval.

MemReader combines MemReader-4B, MemReader-0.6B, a ReAct-style trajectory policy, and a memory-operation tool space to actively judge value, ambiguity, and completeness before writing memories. On LongMemEval, MemReader-4B-GRPO achieves 83.00% Overall vs 83.00% for EverMemOS, with 922 vs 2800 average tokens per extraction.

MemReranker connects Qwen3-Reranker with multi-teacher label generation, BCE pointwise distillation, InfoNCE contrastive fine-tuning, and multi-turn dialogue data engineering to build reasoning-aware memory rerankers. On LongMemEval, MemReranker-4B attains MAP 0.8043 versus 0.7259 for Gemini-3-Flash, while MemReranker-0.6B matches or exceeds GPT-4o-mini at ∼8× lower latency.

MemSecBench links Lifecycle Task Packages, Build-MemSecBench-Case Skill, Lifecycle Evaluation Workflow, Evidence-Based LLM Judging, and a 24-configuration matrix to trace malicious semantics through memory systems. MemSecBench reports 50.3% End-to-End Attack Success Rate and 56.1% Selective Repair Success Rate across 310 cases and 24 configurations, contrasting Native with Mem0, Mem0-Graph, and A-MEM.

Multi-Agent Memory Architecture organizes Agent IO Layer, Agent Cache Layer, Agent Memory Layer, Agent Cache Sharing, and Agent Memory Access Protocol into a computer-architecture-style design for LLM agents. Multi-Agent Memory Architecture’s main result is a conceptual unification of shared and distributed memory plus a research agenda for multi-agent memory consistency instead of benchmark gains.

eTAMP stores attacker-crafted payloads inside Trajectory Memory, then reactivates them via semantic Cross-Site Task Pairing and Chaos Monkey induced frustration during later tasks. On (Visual)WebArena, eTAMP achieves up to 32.5% ASRB on GPT-5-mini and 23.4% on GPT-5.2 under Frustration Exploitation, showing persistent cross-session compromise without direct memory access.

RealMem constructs realistic long-term project dialogues via Project Foundation Construction, Multi-Agent Dialogue Generation, and Memory and Schedule Management over eleven scenarios and 2,000+ cross-session dialogues. On RealMem, the Oracle QA Score reaches 0.804 while the strongest memory system, MemoryOS, achieves only 0.567, quantifying the difficulty of real-world project memory.

TriMem combines dense storage with efficient retrieval, scattered fact with integrated profile, and lifelong evolution with optimized prompt to build a three-level agent memory. On LoCoMo with GPT-4.1-mini, TriMem achieves 54.26 F1 average versus 50.30 F1 for SimpleMem, and up to 57.04 F1 with GPT-5-nano.

Rethinking Memory Mechanisms of Foundation Agents in the Second Half organizes foundation agent memory using a three-dimensional taxonomy of Memory Substrates, Memory Cognitive Mechanisms, and Memory Subjects plus an operation and optimization view. Rethinking Memory Mechanisms of Foundation Agents in the Second Half synthesizes 218 memory-related agent papers from 2023 Q1–2025 Q4, highlighting the sharp acceleration of external memory and working or episodic mechanisms in 2025.

TAME stores experiences as (query, experience, usefulness, trust annotation) in a shared Strategy Memory Bank, orchestrated by an Executor–Evaluator loop with feedback-driven memory evolution. On the GPT-5.2 AIME benchmark, TAME reaches 0.733 accuracy versus 0.587 for ReasoningBank, a +0.146 improvement while preserving trustworthiness on Trust-Memevo.

TA-Mem processes long conversations via an Episodic Memory Constructor, Multi-Indexed Database with Tools, and Memory Retrieval Agent that cooperate to chunk, index, and query structured memory pages. On the LoCoMo dataset, TA-Mem achieves 55.95 F1 and 51.47 BLEU-1 on temporal questions, beating Mem0 and MemoryOS while using 3755 tokens on average.

Weighted Memory Tree organizes agent history into a Hierarchical Memory Tree, Dynamic Retention Scoring, Memory Controller and Lifecycle Operations, and a Utility-Aware Prompt Synthesizer that jointly decide which memories stay active. On GAIA-Text, Weighted Memory Tree reaches 33.86% accuracy with Qwen3-8B versus 20.47% for Linear History, while reducing prompt-token usage from 43.67M to 32.48M.

Dependency-aware Semantic Garbage Collection (DSGC) augments similarity-only retention with one-hop propagation over a supplied prerequisite graph E, using semantic relevance scores ri, propagation term πi, and a greedy retained subset S under budget. On the fixed benchmark with structurally indirect prerequisites, DSGC lifts full-chain retention from 0.03 to 0.90 under the lexical encoder and from 0.23 to 1.00 under the sentence encoder, compared to the similarity-only baseline.

WorldMemArena evaluates multimodal agent memory via an Action World Interaction Loop and a four-stage lifecycle of Observe to Write, Update and Consolidate, Retrieve for Decision, and Use and Act across 461 multi-session tasks. On WorldMemArena, systems like M2A reach 86.83% memory Recall but only 50.14% QA-C, revealing that storing more correct memories does not guarantee better performance compared to baselines like UniversalRAG at 39.62% QA-C.

AgentSafe secures LLM-based multi-agent systems by combining ThreatSieve for permissioned communication and identity checks with HierarCache for hierarchical memory and junk isolation. On RIOH and WCEI, AgentSafe achieves up to 96.88% defense rate in memory-based attacks and maintains CSR above 0.65, compared to baselines dropping below 0.4 without AgentSafe.

A-MEM organizes agent memory via Note Construction, Link Generation, Memory Evolution, and Retrieve Relative Memory to build an evolving, interconnected note graph. On the LoCoMo dataset, A-MEM with GPT-4o-mini reaches 27.02 F1 on Multi Hop questions, +17.87 over ReadAgent, while cutting average token length from 16,910 to 2,520.

A-MemGuard combines consensus-based validation, dual-memory structure, lesson memory, and path divergence scoring to sanitize retrieved memories and revise actions using past failures. On EHRAgent under AgentPoison, A-MemGuard reduces ASR-r from 100.0% to 2.13% and ASR-t from 100.0% to 6.38%, far below LLM Auditor and Distil Classifier.

DAM decomposes agent memory into a Read Policy, Write Policy with Sub-policy, Value Function, Uncertainty Estimator, and an Aggregate Policy that arbitrates storage operations. DAM’s main result is a formal decision-theoretic framework that clarifies the limits of heuristic memory and unifies existing memory-augmented agents under a single uncertainty-aware control perspective.

CAM builds hierarchical schemata using an incremental overlapping clustering algorithm, ego centric disentanglement, and a Prune and Grow associative strategy for retrieval. On NovelQA, CAM achieves 52.3 ACC-L versus RAPTOR’s 47.8, a +4.5 point gain while also improving efficiency in long-text reading comprehension.

Evaluating Long-Term Memory for Long-Context Question Answering compares Full Context, RAG, A-Mem, RAG+PromptOpt, and RAG+EpMem memory components across semantic, episodic, and procedural memory for long conversational QA. On LoCoMo, RAG+EpMem reaches an average F1 ranking of 1.83 for Llama 3.2-3B Instruct and 1.80 for GPT-4o mini while using around 1,000 tokens per query versus over 23,000 for Full Context.

MemoryAgentBench standardizes multi-turn datasets into chunked conversations with memorization prompts, then evaluates long-context agents, RAG agents, and agentic memory agents across Accurate Retrieval, Test-Time Learning, Long-Range Understanding, and Selective Forgetting. On the overall score in Table 3, the GPT-4.1-mini long-context agent reaches 71.8 on Accurate Retrieval tasks compared to 49.2 for the GPT-4o-mini long-context baseline.

MaRS organizes agent memory into episodic, semantic, social, and task nodes with provenance, scored by a privacy-aware retention controller and governed by FIFO, LRU, Priority Decay, Reflection-Summary, Random-Drop, and Hybrid policies. On the FiFA benchmark, the Hybrid policy in MaRS achieves a composite score of ≈0.911 across 300 runs and five memory budgets, outperforming simpler policies while preserving privacy and cost efficiency.

General Agentic Memory (GAM) combines a Memorizer, Researcher, page-store, and memory to keep full trajectories while constructing lightweight guidance for deep research. On RULER 128K retrieval, GAM achieves 97.70% accuracy compared to 94.25% for RAG using GPT-4o-mini, while also reaching 64.07 F1 on HotpotQA-56K.

H-MEM organizes long-term dialogue history into a four-level hierarchy using the Domain Layer, Category Layer, Memory Trace Layer, and Episode Layer, with positional index encoding and a dynamic memory update mechanism. On the LoCoMo dataset, H-MEM raises average F1 by 14.98 and BLEU-1 by 12.77 points over baselines like MemoryBank and MemGPT while reducing retrieval compute from 7.34×10^9 to 4.38×10^7 operations on adversarial tasks.

HINDSIGHT organizes agent memory into four networks via TEMPR and layers CARA on top to retain, recall, and reflect with explicit opinions and behavioral profiles. On LongMemEval, HINDSIGHT with Gemini-3 Pro scores 91.4% overall versus 60.2% for full-context GPT-4o, while HINDSIGHT with OSS-20B jumps from 39.0% to 83.6% over a full-context OSS-20B baseline.

IMDMR combines a Memory Storage Layer, Multi-Dimensional Search Engine, Intelligent Query Processor, and Response Generation Module to retrieve conversational memories across semantic, entity, category, intent, context, and temporal dimensions. On the synthetic 1,000 conversation benchmark, IMDMR-Prod achieves an overall score of 0.792 compared to 0.207 for spaCy + RAG, a 3.8x improvement.

LiCoMemory organizes long term dialogue with CogniGraph, Query Processing and Integrated Rerank, and Real Time Interactions to keep session summaries, triples, and chunks linked. On LongMemEval with GPT-4o-mini, LiCoMemory reaches 73.80% accuracy and 76.63% recall, beating Mem0g by 9.0 and 7.1 points.

MemBench evaluates LLM-based agents with multi-scenario datasets, multi-level memory content, and a time-aware benchmark using components like Multi-scenario Dataset, Multi-level Memory Content, and Multi-metric Evaluation. MemBench shows that mechanisms such as GenerativeAgent, MemGPT, MemoryBank, and SCMemory can drop from accuracies around 0.7 on 10k-token settings to roughly 0.3–0.4 at 100k tokens, exposing clear capacity limits.

MemEvolve decomposes agent memory into Encode, Store, Retrieve, and Manage modules and meta evolves these components via a dual evolution process over candidate architectures. On xBench DeepSearch, MemEvolve with GPT 5 mini raises Flash Searcher pass@1 from 69.0 to 74.0 and WebWalkerQA accuracy from 58.82 to 61.18 while keeping API cost near 0.141 per query.

Memoria augments LLM chats with structured conversation logging, dynamic user persona via KG, session level memory for real time context, and seamless retrieval for context aware responses to provide persistent, interpretable memory. On LongMemEvals single-session-user and knowledge-update subsets, Memoria reaches 87.1% and 80.8% accuracy respectively, surpassing A-Mem (OpenAI) while using much shorter prompts.

MemoriesDB stores each Memory Record, Edges and Relations, and the Temporal Semantic Stack inside PostgreSQL with pgvector, exposing unified temporal–semantic–relational queries. MemoriesDB’s main result is a working implementation that demonstrates scalable time-bounded recall and hybrid semantic–structural queries on commodity SQL infrastructure without specialized vector or graph engines.

Memory in the Age of AI Agents formalizes agent memory with Memory Formation, Memory Evolution, and Memory Retrieval operators, and classifies memories into token-level, parametric, and latent forms plus factual, experiential, and working functions. Memory in the Age of AI Agents’ main result is a unified Forms–Functions–Dynamics framework that consolidates fragmented LLM agent memory work, benchmarks, and open-source frameworks into a coherent taxonomy.

MEMTRACK evaluates long term memory by combining a Containerized Environment, Timeline Generator, Event Parser, and Execution Monitoring with multi platform tools like Slack, Linear, and Git. On the MEMTRACK benchmark, the best performing GPT 5 configuration achieves only a 60% Correctness score, revealing substantial room for improvement over existing memory backends.

PersonaMem-v2 combines PERSONAMEM-V2: IMPLICIT PERSONAS, RL with Long-Context Reasoning, RL with Agentic Memory, and a User Privacy-Aware Design to train Qwen3-4B with GRPO on implicit user preferences from long, noisy histories. PersonaMem-v2 achieves 55.2% MCQ and 60.7% open-ended accuracy on PERSONAMEM-V2, surpassing GPT-5-Chat’s 45.6% and 46.2% while using a 2k-token agentic memory instead of full 32k–128k contexts.

Semantic Anchoring enriches conversational memory by combining a hybrid memory store with dense and symbolic indexes, structured memory representation tuples, hybrid storage and indexing, and a retrieval scoring method. On MultiWOZ-Long, Semantic Anchoring reaches 83.5% Factual Recall and 80.8% Discourse Coherence, beating Entity-RAG by 7.6 and 8.6 points respectively.

MEXTRA crafts black box attacking prompts and automated diverse prompt generators that target the memory module, similarity scoring function, retrieval depth, memory size, and LLM backbone. MEXTRA extracts 50 queries from a 200 record EHRAgent memory and 26 from RAP, with extracted efficiency up to 0.42 compared to weaker baselines without workflow aligned prompts.

WebATLAS combines a Planner, Actor, Critic, and Multi-layered Memory (Working Memory, Cognitive Map, Semantic Memory) to simulate and score actions before executing them on the web. On WebArena-Lite, WebATLAS achieves 63.0% average success versus 53.9% for Plan-and-Act, a +9.1 point gain without website-specific fine-tuning.