All AI memory papers

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).

EMemBench builds on Interaction episode, Benchmark generator, Structured logging, State reconstruction, and Programmatic QA generation to turn each agent’s own game trajectory into verifiable episodic-memory questions. On text-only games, EMemBench shows A-MEM with Qwen3-32B achieves 51.9% Overall ACC compared to 44.9% for the Qwen3-32B in-context baseline (+7.0 points).

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.

ER-MIA combines Content-Based MIAs, Question-Targeted MIAs, and an attack Arsenal of primitive and ensemble strategies to poison similarity-based retrieval in long-term memory systems. On the LoCoMo benchmark, ER-MIA’s ensembles cut Mem0’s overall F1 from 23.60% to 2.87% (-87.6%) and reduce single attacks like Harsh Instruction by up to -71.5% F1.

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%.

FinPerMA combines Persona Synthesis, Event Timeline Construction, the three-layer Impact Model, Dialogue Synthesis, and Evaluation Task Design to generate frozen, event-grounded investor trajectories for memory benchmarking. On the FinPerMA v8gold corpus, full-context Qwen-3.8 reaches 46.9% overall accuracy and 38.7% MCQ, while retrieval-based memory on Qwen3.7-Max recovers ≈88% of the no-memory–full-context gap with only 1.40k tokens of context.

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.

Memora combines Seed Data Design, Session Simulation, Conversation Generation, and Questions and Evaluation Criteria to build weeks-to-months multi-session benchmarks with evolving memories. Using FAMA on Memora, long-term memory agents like MemoBase fall from 43.60 to 15.18 remembering FAMA between weekly and quarterly settings, revealing severe brittleness under heavy memory mutation.

From Storage to Experience formalizes LLM agent memory into Storage, Reflection, and Experience, detailing mechanisms like Linear, Vector, Structured storage and Introspection, Environment, Coordination reflection. From Storage to Experience’s main result is a unified evolutionary framework that connects fragmented operating system style and cognitive science inspired memory designs into a single roadmap.

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.

GradMem combines a WRITE phase, a READ phase, a context encoder Eθ, a self-supervised WRITE objective Lwrite, and a meta-learned initialization M0 to optimize prefix memory tokens via test-time gradient descent while keeping model weights frozen. On associative KV-retrieval with 96 key–value pairs, GradMem with 5 gradient WRITE steps reaches 88.4% exact match versus 12.9% for forward-only RMT with the same 8-vector memory.

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.

GroupMemBench combines Graph-grounded Message Synthesis, Path-Based Context Sampling, Message Generation, Adversarial Question Synthesis, and a Solve–Judge–Refine Loop to stress-test group memory in LLM agents. On GroupMemBench, Hindsight reaches 46.01% average accuracy while a simple BM25 baseline attains 43.22%, revealing that current memory ingestion pipelines fail to preserve crucial multi-user structure.

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.

HiMem builds Episode Memory, Note Memory, Knowledge Alignment, and Memory Reconsolidation to construct, retrieve, and update long-term dialogue memories hierarchically. On the LoCoMo benchmark, HiMem achieves an overall 80.71% GPT-Score, beating the best baseline SeCom at 69.03% by 11.68 points.

HingeMem combines Boundary Guided Long-Term Memory, Dialogue Boundary Extraction, Memory Construction, Query Adaptive Retrieval, Hyperedge Rerank, and Adaptive Stop to segment dialogues into element-indexed hyperedges and plan query-specific retrieval. On LOCOMO, HingeMem achieves 63.9 overall F1 and 75.1 LLM-as-a-Judge score, surpassing the best baseline Zep (56.9 F1) by 7.0 F1 without using category-specific QA formats.

Human-Like Lifelong Memory combines Executive Function and Working Memory, a Memory Service Knowledge Graph, and a Thalamic Gateway to implement dual-process, valence-aware lifelong memory. Human-Like Lifelong Memory is a theoretical framework with seven functional properties and testable predictions rather than benchmark numbers against specific baselines.

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.