arxiv: 2605.13438 · v1 · submitted 2026-05-13 · 💻 cs.AI · cs.CL

Recognition: 2 theorem links

· Lean Theorem

Cognifold: Always-On Proactive Memory via Cognitive Folding

Suli Wang , Yiqun Duan , Yu Deng , Rundong Zhao , Dai Shi , Xinliang Zhou

Authors on Pith no claims yet

Pith reviewed 2026-05-14 19:15 UTC · model grok-4.3

classification 💻 cs.AI cs.CL

keywords cognitive foldingagent memoryproactive cognitiongraph self-organizationcomplementary learning systemsintent emergenceevent stream processingautonomous agents

0 comments

The pith

CogniFold folds event streams into self-organizing graph structures that surface proactive intents at concept density thresholds.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

CogniFold is presented as an always-on memory architecture for agents that continuously processes incoming event streams by folding them into persistent cognitive structures. It extends Complementary Learning Systems theory by adding a prefrontal intent layer and relies on graph-topology rules for assembly, merging, decay, relinking, and intent surfacing. The system becomes proactive when concept-cluster density crosses a threshold, allowing higher-level cognition to emerge from raw events and accumulated knowledge. Evaluation on CogEval-Bench shows the resulting structures align with cognitive expectations, while performance holds on standard memory benchmarks across multiple domains.

Core claim

CogniFold achieves proactive memory by extending Complementary Learning Systems theory to three layers and applying graph-topology self-organization that assembles cognitive structures from fragmented events, merges semantically similar ones, decays stale information, relinks through association, and surfaces intents when concept-cluster density crosses a threshold, bootstrapping progressively higher-level cognition without additional human-specified rules.

What carries the argument

Graph-topology self-organization, which assembles, merges, decays, relinks, and surfaces intents from event streams when concept-cluster density crosses a threshold.

If this is right

Agents bootstrap higher-level cognition directly from incoming event streams through continuous structural folding.
Intents emerge proactively when concept clusters reach density thresholds rather than through explicit retrieval.
Memory structures evolve via assembly, merging, decay, and relinking while preserving performance on conventional benchmarks.
The three-layer extension of Complementary Learning Systems supports intentional control in agent decision-making.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The folding process could enable anticipation of user needs by associating past patterns with new contexts in real time.
Similar self-organization rules might transfer to continual learning settings where event streams include sensor or multimodal data.
The density threshold parameter could determine the balance between over- and under-surfacing of intents in different task environments.

Load-bearing premise

The graph-topology self-organization rules including the density threshold will produce memory structures that genuinely match cognitive expectations and enable proactive behavior without additional human-specified rules or post-hoc tuning.

What would settle it

Running CogniFold on CogEval-Bench and checking whether the generated structures match expected cognitive patterns or fail to surface appropriate intents in scenarios with clear dense concept clusters.

Figures

Figures reproduced from arXiv: 2605.13438 by Dai Shi, Rundong Zhao, Suli Wang, Xinliang Zhou, Yiqun Duan, Yu Deng.

**Figure 1.** Figure 1: From reactive to proactive agent memory. Conventional agents wait for explicit user queries (left) or graft delayed, application-layer triggers onto a reactive memory (middle). In contrast, COGNIFOLD (right) processes unprompted, asynchronous events instantly within its memory substrate, simultaneously reactivating related dormant concepts (e.g., the Vienna hotel and concert). be a property of the memory s… view at source ↗

**Figure 2.** Figure 2: The COGNIFOLD Architecture: Conceptual Bootstrapping via Tri-Layered Cognitive Folding. Extending the Complementary Learning Systems (CLS) framework, the memory substrate continuously metabolizes streaming events through three stages: accumulating raw episodic traces (Hippocampal layer), consolidating redundant patterns into semantic concepts (Neocortical layer), and crystallizing intents (Prefrontal layer… view at source ↗

**Figure 3.** Figure 3: Continuous cognitive metabolism. Under an asynchronous event stream, the memory substrate dynamically self-organizes. The graph autonomously consolidates episodic events (Panel 3), merges associated schemata (Panel 4), and crystallizes goal-directed intents from converging concept density (Panel 5). This living topology natively supports top-down cognitive bias (Panel 7), natural temporal decay (Panel 8), … view at source ↗

**Figure 4.** Figure 4: Downstream benchmarks at a glance. CogniFold (indigo, bold) against the most-cited published baselines for each benchmark, sorted by score with the best on top. Metric varies per benchmark; sample sizes are 500 for MuSiQue, NarrativeQA, MuTual, StreamingQA, ToMi, and 100 for BABILong. benchmark, and against the most-cited published baselines on each of the remaining five; [PITH_FULL_IMAGE:figures/full_fig… view at source ↗

read the original abstract

Existing agent memory remains predominantly reactive and retrieval-based, lacking the capacity to autonomously organize experience into persistent cognitive structure. Toward genuinely autonomous agents, we introduce Cognifold, a brain-inspired "always-on" agent memory designed for the next generation of proactive assistants. CogniFold continuously folds fragmented event streams into self-emerging cognitive structures, bootstrapping progressively higher-level cognition from incoming events and accumulated knowledge. We ground this by extending Complementary Learning Systems (CLS) theory from two layers (hippocampus, neocortex) to three, adding a prefrontal intent layer. Emulating the prefrontal cortex as the locus of intentional control and decision-making, CogniFold achieves this through graph-topology self-organization: cognitive structures proactively assemble under the stream, merge when semantically similar, decay when stale, relink through associative recall, and surface intents when concept-cluster density crosses a threshold. We evaluate structural formation using CogEval-Bench, demonstrating that CogniFold uniquely produces memory structures that match cognitive expectations and concept emergence. Furthermore, across 7 broad-coverage benchmarks spanning five cognitive domains, we validate that CogniFold simultaneously performs robustly on conventional memory benchmarks.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Cognifold proposes a graph-based proactive memory extension to CLS theory, but the self-organization depends on author-defined rules and thresholds.

read the letter

Cognifold introduces a memory architecture for AI agents that aims to be always-on and proactive by folding streams of events into emerging cognitive structures using graph topology rules. This extends the classic two-layer complementary learning systems model by adding a prefrontal intent layer that surfaces goals when concept clusters reach a certain density. The paper does a decent job laying out the mechanism: structures assemble from events, merge on semantic similarity, decay over time, relink through associations, and trigger intents at the threshold. It backs this with results on CogEval-Bench showing structures that align with cognitive expectations, and it holds up across seven benchmarks covering multiple domains. That's a reasonable spread for an initial claim. The soft spot is that the graph self-organization rules, including the density threshold for surfacing intents, are defined upfront by the authors. This means the claimed autonomy in producing cognitive structures may still hinge on those human choices rather than emerging purely. The benchmark results could be influenced by how well the rules fit the test design, and without more on parameter sensitivity or independent validation, it's not clear how general the proactive behavior really is. This paper is for researchers focused on agent memory systems and cognitive modeling in AI. It offers a concrete proposal that builds on solid prior work in a new way. I recommend sending it for peer review. The synthesis is novel enough and the problem important enough that expert input would help clarify the strengths and the remaining gaps in the evidence.

Referee Report

2 major / 1 minor

Summary. The paper introduces Cognifold, a brain-inspired always-on agent memory system that continuously folds fragmented event streams into self-emerging cognitive structures via graph-topology self-organization (assemble, merge, decay, relink, and surface intents at a concept-cluster density threshold). It extends Complementary Learning Systems theory to three layers by adding a prefrontal intent layer and claims that the resulting structures match cognitive expectations on CogEval-Bench while delivering robust performance across 7 benchmarks spanning five cognitive domains.

Significance. If the central claims hold, Cognifold could advance proactive AI agents by enabling autonomous bootstrapping of higher-level cognition from raw events without constant external supervision. The explicit grounding in CLS theory and the dual evaluation on structural matching plus conventional benchmarks would position the work as a concrete bridge between cognitive neuroscience and agent memory design.

major comments (2)

[Abstract] Abstract: the claim that structures 'match cognitive expectations' and 'uniquely' emerge is load-bearing for the central contribution, yet the manuscript provides no equations, pseudocode, or implementation details for the assemble/merge/decay/relink rules or the density threshold, rendering the self-organization claim unverifiable from the text.
[Evaluation] Evaluation section: the reported robustness on 7 benchmarks is stated without error bars, statistical tests, baseline comparisons, or data-exclusion criteria, so it is impossible to determine whether performance stems from the proposed mechanisms or from post-hoc calibration of the free parameter (concept-cluster density threshold).

minor comments (1)

[Abstract] The term 'CogEval-Bench' is introduced without a reference or description of its construction, making it difficult to assess whether the benchmark embeds definitions aligned with the proposed mechanisms.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive feedback. We agree that the current manuscript requires additional technical detail and statistical rigor to fully substantiate the central claims. Below we respond point-by-point to the major comments and outline the revisions we will make.

read point-by-point responses

Referee: [Abstract] Abstract: the claim that structures 'match cognitive expectations' and 'uniquely' emerge is load-bearing for the central contribution, yet the manuscript provides no equations, pseudocode, or implementation details for the assemble/merge/decay/relink rules or the density threshold, rendering the self-organization claim unverifiable from the text.

Authors: We agree that the self-organization rules must be presented with sufficient formality for independent verification. In the revised manuscript we will add an explicit subsection (and corresponding appendix) containing the mathematical definitions of the assemble, merge, decay, and relink operations, the precise density-threshold criterion, and pseudocode for the full graph-topology update cycle. These additions will make the emergence process reproducible from the text alone. revision: yes
Referee: [Evaluation] Evaluation section: the reported robustness on 7 benchmarks is stated without error bars, statistical tests, baseline comparisons, or data-exclusion criteria, so it is impossible to determine whether performance stems from the proposed mechanisms or from post-hoc calibration of the free parameter (concept-cluster density threshold).

Authors: We accept this criticism. The revised evaluation section will report mean performance with standard-error bars across multiple random seeds, include paired statistical tests (e.g., Wilcoxon or t-tests) against the strongest baselines, document all baseline implementations and hyper-parameter choices, and state the data-exclusion criteria. We will also describe the procedure used to select the density threshold (including any sensitivity analysis) to demonstrate that performance is not the result of post-hoc tuning. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper defines CogniFold via explicit graph-topology operations (assemble, merge, decay, relink, and density-threshold surfacing) that extend CLS theory to a prefrontal layer, then evaluates the resulting structures on CogEval-Bench plus seven external benchmarks. No equations, fitted parameters, or self-citations are presented that would make the claimed self-emergent cognitive structures equivalent to the input rules by construction. The mechanisms are author-specified (as is standard for such architectures) but the performance claims rest on independent benchmark outcomes rather than tautological re-labeling or reduction to the same inputs. The derivation therefore remains self-contained against external validation.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The central claim depends on brain-inspired modeling assumptions and a small number of implicit thresholds whose values are not reported in the abstract.

free parameters (1)

concept-cluster density threshold
The density level at which intents are surfaced is introduced as a trigger but no value or fitting procedure is given.

axioms (1)

domain assumption Complementary Learning Systems theory can be extended from two layers to three by adding a prefrontal intent layer that governs intentional control.
The paper grounds the architecture in this extension without providing independent justification or formal mapping to neuroscientific data.

invented entities (1)

prefrontal intent layer no independent evidence
purpose: To serve as the locus of intentional control and decision-making that surfaces intents from concept clusters.
This is a new postulated layer added to the CLS framework; no independent evidence such as predicted neural correlates is supplied.

pith-pipeline@v0.9.0 · 5507 in / 1399 out tokens · 59260 ms · 2026-05-14T19:15:48.422416+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

surface intents when concept-cluster density crosses a threshold... graph-topology self-organization: cognitive structures proactively assemble under the stream, merge when semantically similar, decay when stale, relink through associative recall
IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

extending Complementary Learning Systems (CLS) theory from two layers... to three, adding a prefrontal intent layer

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

77 extracted references · 17 canonical work pages · 11 internal anchors

[1]

Cambridge university press, 1995

Frederic Charles Bartlett.Remembering: A study in experimental and social psychology. Cambridge university press, 1995

1995
[2]

Titans: Learning to Memorize at Test Time

Ali Behrouz, Peilin Zhong, and Vahab Mirrokni. Titans: Learning to memorize at test time. arXiv preprint arXiv:2501.00663, 2024

work page internal anchor Pith review arXiv 2024
[3]

Time- dependent reorganization of brain circuitry underlying long-term memory storage.Nature, 400 (6745):671–675, 1999

Bruno Bontempi, Catherine Laurent-Demir, Claude Destrade, and Robert Jaffard. Time- dependent reorganization of brain circuitry underlying long-term memory storage.Nature, 400 (6745):671–675, 1999

1999
[4]

Harvard University Press, 1987

Michael Bratman.Intention, plans, and practical reason. Harvard University Press, 1987

1987
[5]

The origin of concepts.Journal of Cognition and Development, 1(1):37–41, 2000

Susan Carey. The origin of concepts.Journal of Cognition and Development, 1(1):37–41, 2000

2000
[6]

Mem0: Building Production-Ready AI Agents with Scalable Long-Term Memory

Prateek Chhikara, Dev Khant, Saket Aryan, Taranjeet Singh, and Deshraj Yadav. Mem0: Building production-ready ai agents with scalable long-term memory.arXiv preprint arXiv:2504.19413, 2025

work page internal anchor Pith review Pith/arXiv arXiv 2025
[7]

Whatever next? predictive brains, situated agents, and the future of cognitive science.Behavioral and brain sciences, 36(3):181–204, 2013

Andy Clark. Whatever next? predictive brains, situated agents, and the future of cognitive science.Behavioral and brain sciences, 36(3):181–204, 2013

2013
[8]

Mutual: A dataset for multi- turn dialogue reasoning

Leyang Cui, Yu Wu, Shujie Liu, Yue Zhang, and Ming Zhou. Mutual: A dataset for multi- turn dialogue reasoning. InProceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 1406–1416, 2020

2020
[9]

The prefrontal cortex controls memory organization in the hippocampus.Nature Neuroscience, pages 1–12, 2026

André F de Sousa, Zachary E Zeidler, Daniel G Almeida-Filho, Yang Shen, Alessandro Luchetti, Shana Simanian, Mouaz Mardini, Laura A DeNardo, and Alcino J Silva. The prefrontal cortex controls memory organization in the hippocampus.Nature Neuroscience, pages 1–12, 2026

2026
[10]

[image] memory: A contribution to experimental psychology.Annals of neurosciences, 20(4):155, 2013

Hermann Ebbinghaus. [image] memory: A contribution to experimental psychology.Annals of neurosciences, 20(4):155, 2013

2013
[11]

From Local to Global: A Graph RAG Approach to Query-Focused Summarization

Darren Edge, Ha Trinh, Newman Cheng, Joshua Bradley, Alex Chao, Apurva Mody, Steven Truitt, Dasha Metropolitansky, Robert Osazuwa Ness, and Jonathan Larson. From local to global: A graph rag approach to query-focused summarization.arXiv preprint arXiv:2404.16130, 2024

work page internal anchor Pith review Pith/arXiv arXiv 2024
[12]

Memory: organization and control.Annual review of psychology, 68: 19–45, 2017

Howard Eichenbaum. Memory: organization and control.Annual review of psychology, 68: 19–45, 2017

2017
[13]

Prospective memory: Multiple retrieval processes

Gilles O Einstein and Mark A McDaniel. Prospective memory: Multiple retrieval processes. Current Directions in Psychological Science, 14(6):286–290, 2005

2005
[14]

Learning and development in neural networks: The importance of starting small.Cognition, 48(1):71–99, 1993

Jeffrey L Elman. Learning and development in neural networks: The importance of starting small.Cognition, 48(1):71–99, 1993

1993
[15]

The organization of recent and remote memories

Paul W Frankland and Bruno Bontempi. The organization of recent and remote memories. Nature reviews neuroscience, 6(2):119–130, 2005

2005
[16]

The free-energy principle: a unified brain theory?Nature reviews neuroscience, 11(2):127–138, 2010

Karl Friston. The free-energy principle: a unified brain theory?Nature reviews neuroscience, 11(2):127–138, 2010

2010
[17]

Neurobiology of schemas and schema-mediated memory

Asaf Gilboa and Hannah Marlatte. Neurobiology of schemas and schema-mediated memory. Trends in cognitive sciences, 21(8):618–631, 2017

2017
[18]

LightRAG: Simple and Fast Retrieval-Augmented Generation

Zirui Guo, Lianghao Xia, Yanhua Yu, Tian Ao, and Chao Huang. Lightrag: Simple and fast retrieval-augmented generation.arXiv preprint arXiv:2410.05779, 2(3), 2024

work page internal anchor Pith review Pith/arXiv arXiv 2024
[19]

Hipporag: Neurobio- logically inspired long-term memory for large language models.Advances in neural information processing systems, 37:59532–59569, 2024

Bernal J Gutiérrez, Yiheng Shu, Yu Gu, Michihiro Yasunaga, and Yu Su. Hipporag: Neurobio- logically inspired long-term memory for large language models.Advances in neural information processing systems, 37:59532–59569, 2024

2024
[20]

From RAG to memory: Non- parametric continual learning for large language models.CoRR, abs/2502.14802, 2025

Bernal Jiménez Gutiérrez, Yiheng Shu, Weijian Qi, Sizhe Zhou, and Yu Su. From rag to memory: Non-parametric continual learning for large language models.arXiv preprint arXiv:2502.14802, 2025

work page arXiv 2025
[21]

Psychology press, 2005

Donald Olding Hebb.The organization of behavior: A neuropsychological theory. Psychology press, 2005

2005
[22]

Evermemos: A self-organizing memory operating system for structured long-horizon reasoning.arXiv preprint arXiv:2601.02163, 2026

Chuanrui Hu, Xingze Gao, Zuyi Zhou, Dannong Xu, Yi Bai, Xintong Li, Hui Zhang, Tong Li, Chong Zhang, Lidong Bing, and Yafeng Deng. Evermemos: A self-organizing memory operating system for structured long-horizon reasoning.arXiv preprint arXiv:2601.02163, 2026. 12

work page arXiv 2026
[23]

MAGMA: A Multi-Graph based Agentic Memory Architecture for AI Agents

Dongming Jiang, Yi Li, Guanpeng Li, and Bingzhe Li. Magma: A multi-graph based agentic memory architecture for ai agents.arXiv preprint arXiv:2601.03236, 2026

work page internal anchor Pith review Pith/arXiv arXiv 2026
[24]

The narrativeqa reading comprehension challenge.Transac- tions of the Association for Computational Linguistics, 6:317–328, 2018

Tomáš Koˇcisk`y, Jonathan Schwarz, Phil Blunsom, Chris Dyer, Karl Moritz Hermann, Gábor Melis, and Edward Grefenstette. The narrativeqa reading comprehension challenge.Transac- tions of the Association for Computational Linguistics, 6:317–328, 2018

2018
[25]

The hungarian method for the assignment problem.Naval research logistics quarterly, 2(1-2):83–97, 1955

Harold W Kuhn. The hungarian method for the assignment problem.Naval research logistics quarterly, 2(1-2):83–97, 1955

1955
[26]

What learning systems do intelligent agents need? complementary learning systems theory updated.Trends in cognitive sciences, 20(7):512–534, 2016

Dharshan Kumaran, Demis Hassabis, and James L McClelland. What learning systems do intelligent agents need? complementary learning systems theory updated.Trends in cognitive sciences, 20(7):512–534, 2016

2016
[27]

Babilong: Testing the limits of llms with long context reasoning-in-a-haystack

Yuri Kuratov, Aydar Bulatov, Petr Anokhin, Ivan Rodkin, Dmitry Sorokin, Artyom Sorokin, and Mikhail Burtsev. Babilong: Testing the limits of llms with long context reasoning-in-a-haystack. Advances in Neural Information Processing Systems, 37:106519–106554, 2024

2024
[28]

Segmentation in the perception and memory of events.Trends in cognitive sciences, 12(2):72–79, 2008

Christopher A Kurby and Jeffrey M Zacks. Segmentation in the perception and memory of events.Trends in cognitive sciences, 12(2):72–79, 2008

2008
[29]

Revisiting the evaluation of theory of mind through question answering

Matthew Le, Y-Lan Boureau, and Maximilian Nickel. Revisiting the evaluation of theory of mind through question answering. InProceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 5872–5877, Hong Kong, China, November

2019
[30]

Revisiting the evaluation of theory of mind through question answering

Association for Computational Linguistics. doi: 10.18653/v1/D19-1598. URL https: //www.aclweb.org/anthology/D19-1598

work page doi:10.18653/v1/d19-1598
[31]

MemOS: A Memory OS for AI System

Zhiyu Li, Chenyang Xi, Chunyu Li, Ding Chen, Boyu Chen, Shichao Song, Simin Niu, Hanyu Wang, Jiawei Yang, Chen Tang, et al. Memos: A memory os for ai system.arXiv preprint arXiv:2507.03724, 2025

work page internal anchor Pith review arXiv 2025
[32]

Stream- ingqa: A benchmark for adaptation to new knowledge over time in question answering models

Adam Liska, Tomas Kocisky, Elena Gribovskaya, Tayfun Terzi, Eren Sezener, Devang Agrawal, Cyprien De Masson D’Autume, Tim Scholtes, Manzil Zaheer, Susannah Young, et al. Stream- ingqa: A benchmark for adaptation to new knowledge over time in question answering models. InInternational Conference on Machine Learning, pages 13604–13622. PMLR, 2022

2022
[33]

Evaluating very long-term conversational memory of llm agents

Adyasha Maharana, Dong-Ho Lee, Sergey Tulyakov, Mohit Bansal, Francesco Barbieri, and Yuwei Fang. Evaluating very long-term conversational memory of llm agents. InProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 13851–13870, 2024

2024
[34]

Clin: A continually learning language agent for rapid task adaptation and generalization

Bodhisattwa Prasad Majumder, Bhavana Dalvi Mishra, Peter Jansen, Oyvind Tafjord, Niket Tandon, Li Zhang, Chris Callison-Burch, and Peter Clark. Clin: A continually learning language agent for rapid task adaptation and generalization.arXiv preprint arXiv:2310.10134, 2023

work page arXiv 2023
[35]

Simple memory: a theory for archicortex.Philosophical Transactions of the Royal Society of London

David Marr. Simple memory: a theory for archicortex.Philosophical Transactions of the Royal Society of London. B, Biological Sciences, 262(841):23–81, 1971

1971
[36]

James L McClelland, Bruce L McNaughton, and Randall C O’Reilly. Why there are comple- mentary learning systems in the hippocampus and neocortex: insights from the successes and failures of connectionist models of learning and memory.Psychological review, 102(3):419, 1995

1995
[37]

Catastrophic interference in connectionist networks: The sequential learning problem

Michael McCloskey and Neal J Cohen. Catastrophic interference in connectionist networks: The sequential learning problem. InPsychology of learning and motivation, volume 24, pages 109–165. Elsevier, 1989

1989
[38]

Memobase: User profile-based long-term memory for AI chatbot applications

MemoDB Team. Memobase: User profile-based long-term memory for AI chatbot applications. https://github.com/memodb-io/memobase, 2026. Version 0.0.18

2026
[39]

The magical number seven, plus or minus two: Some limits on our capacity for processing information.Psychological review, 63(2):81, 1956

George A Miller. The magical number seven, plus or minus two: Some limits on our capacity for processing information.Psychological review, 63(2):81, 1956

1956
[40]

MemU: A memory operating system for agents

NevaMind AI. MemU: A memory operating system for agents. https://github.com/Nev aMind-AI/memU, 2025

2025
[41]

Modularity and community structure in networks.Proceedings of the national academy of sciences, 103(23):8577–8582, 2006

Mark EJ Newman. Modularity and community structure in networks.Proceedings of the national academy of sciences, 103(23):8577–8582, 2006. 13

2006
[42]

Hippocampal conjunctive encoding, storage, and recall: Avoiding a trade-off.Hippocampus, 4(6):661–682, 1994

Randall C O’Reilly and James L McClelland. Hippocampal conjunctive encoding, storage, and recall: Avoiding a trade-off.Hippocampus, 4(6):661–682, 1994

1994
[43]

MemGPT: Towards LLMs as Operating Systems

Charles Packer, Vivian Fang, Shishir_G Patil, Kevin Lin, Sarah Wooders, and Joseph_E Gonza- lez. Memgpt: towards llms as operating systems.arXiv preprint arXiv:2310.08560, 2023

work page internal anchor Pith review Pith/arXiv arXiv 2023
[44]

Generative agents: Interactive simulacra of human behavior

Joon Sung Park, Joseph O’Brien, Carrie Jun Cai, Meredith Ringel Morris, Percy Liang, and Michael S Bernstein. Generative agents: Interactive simulacra of human behavior. InProceed- ings of the 36th annual acm symposium on user interface software and technology, pages 1–22, 2023

2023
[45]

Engram: Effective, lightweight memory orchestration for conversational agents.arXiv preprint arXiv:2511.12960, 2025

Daivik Patel and Shrenik Patel. Engram: Effective, lightweight memory orchestration for conversational agents.arXiv preprint arXiv:2511.12960, 2025

work page arXiv 2025
[46]

Where do you know what you know? the representation of semantic knowledge in the human brain.Nature reviews neuroscience, 8 (12):976–987, 2007

Karalyn Patterson, Peter J Nestor, and Timothy T Rogers. Where do you know what you know? the representation of semantic knowledge in the human brain.Nature reviews neuroscience, 8 (12):976–987, 2007

2007
[47]

Auditing LoCoMo: 6.4% answer-key error rate, judge leniency, and reproducibil- ity failures in long-term conversational memory benchmarks

Penfield Labs. Auditing LoCoMo: 6.4% answer-key error rate, judge leniency, and reproducibil- ity failures in long-term conversational memory benchmarks. https://github.com/dial4 81/locomo-audit, 2026

2026
[48]

Interplay of hippocampus and prefrontal cortex in memory.Current biology, 23(17):R764–R773, 2013

Alison R Preston and Howard Eichenbaum. Interplay of hippocampus and prefrontal cortex in memory.Current biology, 23(17):R764–R773, 2013

2013
[49]

Zep: A Temporal Knowledge Graph Architecture for Agent Memory

Preston Rasmussen, Pavlo Paliychuk, Travis Beauvais, Jack Ryan, and Daniel Chalef. Zep: a temporal knowledge graph architecture for agent memory.arXiv preprint arXiv:2501.13956, 2025

work page internal anchor Pith review Pith/arXiv arXiv 2025
[50]

Policyrag: Prompt-guided symbolic graph memory for interpretable multi-hop retrieval

Tejas Sarnaik, Manan Shah, and Ravi Hegde. Policyrag: Prompt-guided symbolic graph memory for interpretable multi-hop retrieval. https://openreview.net/forum?id=0xlI09pvBs , 2025

2025
[51]

Raptor: Recursive abstractive processing for tree-organized retrieval

Parth Sarthi, Salman Abdullah, Aditi Tuli, Shubh Khanna, Anna Goldie, and Christopher D Manning. Raptor: Recursive abstractive processing for tree-organized retrieval. InThe Twelfth International Conference on Learning Representations, 2024

2024
[52]

Reflexion: Language agents with verbal reinforcement learning.Advances in neural information processing systems, 36:8634–8652, 2023

Noah Shinn, Federico Cassano, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. Reflexion: Language agents with verbal reinforcement learning.Advances in neural information processing systems, 36:8634–8652, 2023

2023
[53]

Memory and the hippocampus: a synthesis from findings with rats, monkeys, and humans.Psychological review, 99(2):195, 1992

Larry R Squire. Memory and the hippocampus: a synthesis from findings with rats, monkeys, and humans.Psychological review, 99(2):195, 1992

1992
[54]

Sleep-dependent memory consolidation and reconsoli- dation.Sleep medicine, 8(4):331–343, 2007

Robert Stickgold and Matthew P Walker. Sleep-dependent memory consolidation and reconsoli- dation.Sleep medicine, 8(4):331–343, 2007

2007
[55]

Cognitive architectures for language agents.Transactions on Machine Learning Research, 2023

Theodore Sumers, Shunyu Yao, Karthik R Narasimhan, and Thomas L Griffiths. Cognitive architectures for language agents.Transactions on Machine Learning Research, 2023

2023
[56]

Supermemory: State-of-the-art memory and context engine for ai

Supermemory Team. Supermemory: State-of-the-art memory and context engine for ai. https: //github.com/supermemoryai/supermemory, 2026

2026
[57]

cognee: Memory control plane for ai agents

Topoteretes. cognee: Memory control plane for ai agents. https://github.com/topoter etes/cognee, 2026

2026
[58]

♭♭ musique: Multihop questions via single-hop question composition.Transactions of the Association for Computational Linguistics, 10:539–554, 2022

Harsh Trivedi, Niranjan Balasubramanian, Tushar Khot, and Ashish Sabharwal. ♭♭ musique: Multihop questions via single-hop question composition.Transactions of the Association for Computational Linguistics, 10:539–554, 2022

2022
[59]

Schemas and memory consolidation

Dorothy Tse, Rosamund F Langston, Masaki Kakeyama, Ingrid Bethus, Patrick A Spooner, Emma R Wood, Menno P Witter, and Richard GM Morris. Schemas and memory consolidation. Science, 316(5821):76–82, 2007

2007
[60]

Episodic and semantic memory.Organization of memory, 1(381-403):1, 1972

Endel Tulving et al. Episodic and semantic memory.Organization of memory, 1(381-403):1, 1972

1972
[61]

How schema and novelty augment memory formation.Trends in neurosciences, 35(4):211–219, 2012

Marlieke TR Van Kesteren, Dirk J Ruiter, Guillén Fernández, and Richard N Henson. How schema and novelty augment memory formation.Trends in neurosciences, 35(4):211–219, 2012. 14

2012
[62]

Voyager: An Open-Ended Embodied Agent with Large Language Models

Guanzhi Wang, Yuqi Xie, Yunfan Jiang, Ajay Mandlekar, Chaowei Xiao, Yuke Zhu, Linxi Fan, and Anima Anandkumar. V oyager: An open-ended embodied agent with large language models. arXiv preprint arXiv:2305.16291, 2023

work page internal anchor Pith review Pith/arXiv arXiv 2023
[63]

MIRIX: Multi-Agent Memory System for LLM-Based Agents

Yu Wang and Xi Chen. Mirix: Multi-agent memory system for llm-based agents.arXiv preprint arXiv:2507.07957, 2025

work page internal anchor Pith review arXiv 2025
[64]

A-MEM: Agentic Memory for LLM Agents

Wujiang Xu, Zujie Liang, Kai Mei, Hang Gao, Juntao Tan, and Yongfeng Zhang. A-mem: Agentic memory for llm agents.arXiv preprint arXiv:2502.12110, 2025

work page internal anchor Pith review Pith/arXiv arXiv 2025
[65]

Swe-agent: Agent-computer interfaces enable automated software engineering.Advances in Neural Information Processing Systems, 37:50528–50652, 2024

John Yang, Carlos E Jimenez, Alexander Wettig, Kilian Lieret, Shunyu Yao, Karthik Narasimhan, and Ofir Press. Swe-agent: Agent-computer interfaces enable automated software engineering.Advances in Neural Information Processing Systems, 37:50528–50652, 2024

2024
[66]

Event perception and memory.Annual review of psychology, 71(1):165–191, 2020

Jeffrey M Zacks. Event perception and memory.Annual review of psychology, 71(1):165–191, 2020

2020
[67]

A model of conceptual bootstrapping in human cognition.Nature Human Behaviour, 8(1):125–136, 2024

Bonan Zhao, Christopher G Lucas, and Neil R Bramley. A model of conceptual bootstrapping in human cognition.Nature Human Behaviour, 8(1):125–136, 2024

2024
[68]

Judging llm-as-a-judge with mt-bench and chatbot arena.Advances in neural information processing systems, 36:46595–46623, 2023

Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Siyuan Zhuang, Zhanghao Wu, Yonghao Zhuang, Zi Lin, Zhuohan Li, Dacheng Li, Eric Xing, et al. Judging llm-as-a-judge with mt-bench and chatbot arena.Advances in neural information processing systems, 36:46595–46623, 2023

2023
[69]

Memorybank: Enhancing large language models with long-term memory

Wanjun Zhong, Lianghong Guo, Qiqi Gao, He Ye, and Yanlin Wang. Memorybank: Enhancing large language models with long-term memory. InProceedings of the AAAI conference on artificial intelligence, volume 38, pages 19724–19731, 2024. 15 Table 6:Design-space for agent memory under continuous streams.Four orthogonal axes distinguish always-on proactive memory ...

2024
[70]

Create concepts for recurring patterns (3+ events)
[71]

Link every concept to grounding events
[72]

Merge near-duplicate concepts via MERGE_NODES
[73]

Create intents only when patterns suggest unmet goals with supporting evidence
[74]

—” = not addressed; “LLM rewrite

Self-review: check for missing edges between concepts that share grounding events The full prompt includes 20 composable sections (edge types, connectivity rules, validation checklist, deduplication, self-review). Domain-specific Y AML profiles override the role section while retaining structural sections. F CogEval-Bench Details F.1 Gold Graph Schemas Fo...

2024
[75]

MEANINGFULNESS: Is this a semantically coherent concept? (1.0 = clearly defined theme/pattern, 0.0 = incoherent or trivial)
[76]

GROUNDEDNESS: Is the concept well-supported by its grounding events? (1.0 = strong evidence, 0.0 = no supporting evidence)
[77]

meaningfulness

ABSTRACTION LEVEL: Is this the right level of abstraction? (1.0 = useful generalization, 0.0 = too specific or too vague) Output JSON: {"meaningfulness": X, "groundedness": X, "abstraction": X} F.13 Embedding Similarity Threshold Sensitivity The Gold F1 computation uses cosine similarity ≥0.75 for concept matching. Table 15 reports COGNIFOLD’s Harmony sco...

work page arXiv