All AI memory papers

MAPLE orchestrates a central Agent with dedicated Memory, Learning, and Personalization sub-agents that run on different timescales but share explicit user models. On the MAPLE-Personas benchmark, MAPLE achieves a mean judge personalization score of 4.78 vs 4.17 for the Baseline and raises trait incorporation from 45% to 75%.

MeClear evaluates retrieved memories via Local Counterfactual Screening, Counterfactual Cooperative Attribution, and Verified Query-Scoped Clearance to decide which records to suppress per query. On ten LoCoMo-based long dialogue memory pools, MeClear attains 85.9% target recall and 82.3% binary task recovery, a +25.5 percentage point gain over Leave-One-Out.

MedMemoryBench builds a four stage pipeline with Patient Profile Construction, Disease Progression Event Generation, Multi turn Sessions Simulation, and Memory Extraction and Query Construction to synthesize long horizon, clinically grounded interactions. On MedMemoryBench Efficient vs Mixed, methods like Letta drop from 51.21% to 41.55% average accuracy, quantifying how memory saturation harms personalized healthcare agents.

MELD merges distributed wiki brains using a merge decision procedure, Patch, status CRDT, and semantic publish subscribe over a typed-link memory substrate. On HotpotQA distractor, MELD’s decentralized merge achieves recall@5 = 0.630 vs 0.619 for a centralized store and 0.595 for naive union, with about 11% less live storage than union.

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.

MemAgent coordinates a Memory Agent, content aware routing architecture, training data synthesis pipeline, InsightGraph, and a short term memory provider to manage heterogeneous memory providers for a task agent. On GAIA, WebWalkerQA, and xBench DS, MemAgent reaches 75.7% average accuracy versus 65.7% for No Memory, a +10.0 percentage point gain.

MemArbiter decomposes interaction histories into atomic memory items, organizes them into Memory Banks, and arbitrates their Dual-Band Memory Representation, Decision-Relevance Signals, and Temporal Presentation Gate before Prompt Assembly. On ALFWorld, MemArbiter achieves 92.54% SR@50 under a 750-token budget, beating Flat Retrieval by 25.38 percentage points.

MEMARENA builds an ego-centric, activity-dense benchmark using MASIM world simulation, PP sessions, PA sessions, ego-centric projection, and benchmark question generation to stress-test personal memory assistants. On MEMARENA-L, MEMARENA shows that at Qwen3-0.6B, switching from Memobase to MemSearch improves Recall from 23.7 to 56.2 (+32.5 pp) and Reasoning from 22.2 to 41.4 (+19.2 pp), while memory search adds only 87/7/48 ms overhead for BM25-RAG/Memobase/MemSearch on a Spark GB10 node.

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.

MemBuilder trains a Qwen3-4B-based agent to coordinate Core Memory, Episodic Memory, Semantic Memory, and Procedural Memory using dense session-level rewards and contribution-aware gradient weighting. On LoCoMo, MemBuilder achieves 84.23% accuracy, beating MIRIX at 77.48% and Claude 4.5 Sonnet memory construction at 82.61%.

MemCalib combines a 15,000-example benchmark, rubric-guided LLM-as-a-Judge evaluation, and the MemCalib-RL training algorithm built on ordered reward channels, bidirectional counterfactual localization, and channel-wise advantage redistribution. On the MemCalib test set, MemCalib-RL raises Sample Calibration Score to 81.12 on Qwen3.5-35B-A3B, a 1.73-point gain over GDPO while also improving Exact Calibration by 2.65 points.

MemConflict builds structured user profiles, simulates long-horizon timelines, injects conflicts, and evaluates systems via black-box Answer Accuracy and white-box Support Evidence Hit and Support Rank Score. On the MemConflict benchmark, MemOS attains 0.5539 average Answer Accuracy, 0.6710 SEH@3, and 0.5879 SRS, revealing large gaps versus systems like LangMem at 0.2822 AA.

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.

MemForest compresses agent memories by combining EventTree Semantic-Temporal Partitioning, EventTree Progressive Merging, and Anchor-Guided Propagation Retrieval into a unified EventTree-based memory forest. MemForest retains 97.1% of Mem0 performance and 99.7% of M3-Agent performance at 50% compression on LoCoMo, LongMemEval, PersonaMem, M3-Bench-robot, and M3-Bench-web while speeding up retrieval by up to 2.24× over Mem0 and M3-Agent.

Mem-Gallery organizes multi-session, multimodal conversations and tasks via Benchmark Construction, a unified Conversational Environment for Memory, and an Evaluation Framework spanning extraction, reasoning, and knowledge management. On Mem-Gallery, MuRAG reaches 0.6966 overall F1 and 0.8856 LLM-Judge on visual-centric search, surpassing NaiveRAG by 0.0992 overall F1.

MemGround evaluates long-term memory via Surface State Memory, Temporal Associative Memory, and Reasoning-Based Memory in TRPG, No Case Should Remain Unsolved, and Type Help scenarios. MemGround’s main result shows frontier GPT-5.2 reaches 51.51% QA Overall in TRPG but only 23.61% QA Overall in Type Help, revealing severe reasoning degradation versus closed-source baselines like Gemini-3-Pro-Preview.

MemGuard governs agent memory using Verifier-Guided Admission, Structured Memory Units, Precision-Oriented Retrieval, and Memory Consolidation and Budgeted Governance. On WebArena with Qwen-3.5-Plus, MemGuard reaches 58.4% SR vs 50.5% for ReasoningBank (+7.9 points).

MEMGUARD decomposes conversations into type-specific memory atoms via Type-Aware Knowledge Decomposition, Self-Verified Extraction, Relational Knowledge Graph, and Query-Adaptive Type Routing stored in type-isolated memory. On HaluMem, MEMGUARD attains 89.53% anti-hallucination accuracy and 94.15 F1, improving anti-hallucination by 28.27 percentage points over MemOS.

MemGym wires environments like τ2-bench, SWE-Gym, WebArena-Infinity, MEMGYM-DR, and MEMGYM-CODEQA through a shared BaseMemoryManager, BaseAgent, BaseRunner, and MEMRM reward model. MemGym’s MEMRM achieves AUROC 0.985 on SWE-Gym compression events, turning multi-minute Docker rollouts into sub-second scalar memory-quality scores.

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.