Weighted Memory Tree: Remembering What Matters for Long-Horizon LLM Agents

Quang Dao; Purvi Kathalkar; Kenneth Eaton

cs.AIAug 21, 2026

Weighted Memory Tree: Remembering What Matters for Long-Horizon LLM Agents

Authors: Quang Dao, Purvi Kathalkar, Kenneth Eaton

Organizations: Rose-Hulman Institute of Technology · Georgia Institute of Technology · Georgia Tech Research Institute

Abstract

Large language model (LLM) agents have demonstrated the ability to solve multi-step tasks requiring planning, tool use, and external information access, yet growing execution histories increase inference cost and expose reasoning to outdated, irrelevant, or misleading information, potentially degrading reasoning quality. Existing memory approaches organize or compress execution histories but provide limited mechanisms for deciding which memories remain active. We introduce the, a hierarchical memory system that organizes execution into tasks, subtasks, and actions while assigning each memory a dynamic retention score. Event-based updates and selection-based decay revise these scores, allowing WMT to preserve useful information, fold completed trajectories, suppress low-utility content, and retain access to folded context. We evaluate WMT on GAIA-Text using Qwen3-8B, Gemma 4 E4B, and Llama-3.1-8B, with ablations and memory-poisoning experiments. Relative to linear memory, WMT improves accuracy by an average of 9.97 percentage points while reducing prompt-token usage by 32.8%. Memory-poisoning experiments show that WMT limits the persistence and propagation of unreliable information. Our results suggest that effective long-horizon agent memory depends less on storing more information than on deciding which information should remain active.

Figures & tables

Figure 1: Status-driven workflow of WMT . Agent interactions update node scores and branch priorities; lifecycle transitions and prompt-selection feedback determine which memories remain active.

Figure 2: A simplified memory-poisoning scenario containing ordinary task and action memories together with intentionally poisoned memories shown in red.

GAIA-Text
Method	Qwen3-8B		Gemma 4 E4B		Llama-3.1-8B
	Acc. (%) $\uparrow$	Tokens (M) $\downarrow$	Acc. (%) $\uparrow$	Tokens (M) $\downarrow$	Acc. (%) $\uparrow$	Tokens (M) $\downarrow$
No Memory	9.45	0.20	14.17	0.13	6.30	0.03
Linear History	20.47	43.67	24.41	65.27	9.45	73.96
A1: Unweighted Tree	15.75	39.24	25.20	53.90	9.45	77.13
Full WMT	33.86	32.48	33.07	37.85	17.32	51.17

Table 1: Performance on GAIA-Text and GAIA across three base models. Accuracy is the percentage of correctly completed tasks, and token usage is the total number of prompt tokens in millions. Bold values indicate the highest accuracy and lowest token usage among memory-based methods for each model and benchmark.

Appendix figures & tables3 assets

Supplementary material from the paper’s appendix.

Appendix

Memory Configuration	Qwen3-8B		Gemma 4 E4B		Llama-3.1-8B
	Acc. (%) $\uparrow$	Tokens (M) $\downarrow$	Acc. (%) $\uparrow$	Tokens (M) $\downarrow$	Acc. (%) $\uparrow$	Tokens (M) $\downarrow$
Tree Memory	15.75	39.24	25.20	53.90	9.45	77.13
Tree + Selection	22.05	74.37	25.20	92.02	14.96	101.22
Tree + Summary	15.75	34.00	25.98	28.58	19.69	66.83
Tree + Selection + Summary	18.90	47.33	29.13	31.55	17.32	77.05
Full WMT	33.86	32.48	33.07	37.85	17.32	51.17

Appendix

Table 2: Component-ablation results on GAIA-Text. Accuracy is the percentage of correctly completed tasks, and prompt-token usage is reported in millions. Bold values indicate the best result for each model and metric among the reported tree-based configurations.

Memory Configuration	Qwen3-8B		Gemma 4 E4B		Llama-3.1-8B
	Acc. (%) $\uparrow$	Tokens (M) $\downarrow$	Acc. (%) $\uparrow$	Tokens (M) $\downarrow$	Acc. (%) $\uparrow$	Tokens (M) $\downarrow$
Tree Memory	18.18	58.46	25.45	65.09	11.52	101.91
Tree + Selection	21.21	100.51	26.67	105.57	16.36	134.15
Tree + Summary	15.15	49.83	24.85	36.02	16.97	94.25
Tree + Selection + Summary	15.76	66.84	28.48	42.68	20.00	112.44
Full WMT	29.70	42.69	30.91	46.64	24.85	69.45

Appendix

Table 3: Component-ablation results on GAIA. Accuracy is the percentage of correctly completed tasks, and prompt-token usage is reported in millions. Bold values indicate the best result for each model and metric among the reported tree-based configurations.

Metric	Linear Memory	Unweighted Tree (A1)	No Memory Controller (A2)	No Semantic Retrieval (A3)	Full WMT
Attack Success Rate $\downarrow$	0.995	0.631	0.601	0.680	0.419
Poison Retrieval Rate $\downarrow$	0.246	0.125	0.158	0.139	0.097
Infection Persistence $\downarrow$	1.000	0.009	1.000	0.009	0.009
Blast Radius $\downarrow$	0.906	0.388	0.509	0.669	0.315
Amplification Factor $\downarrow$	6.103	1.280	1.638	1.977	0.965
Task Success Rate $\uparrow$	0.183	0.431	0.451	0.393	0.575

Appendix

Table 4: Memory-poisoning results for WMT and its ablations. Arrows indicate the preferred direction. Bold values indicate the best result for each metric; ties are bolded jointly. Linear Memory retains the full sequential history, A1 removes weighting, A2 removes lifecycle control, and A3 removes semantic memory selection.

LLM Agent Memory

Large language model agent memory lets agents retain, organize, retrieve, and update information across interactions and long-running tasks.

See all 357 papers

ActiveMem: Dynamic Latent Memory Trees for Long-Horizon Agents
Sep 27, 2026Song-Li Wu, Jingyi Wang, Zhaocheng Du +1LLM Agent Memory Memory-Augmented Language Model Agents
Tsinghua University · Huawei Noah’s Ark Lab
LeanMem: Simple and Efficient Long-Term Memory for LLM Agents
Aug 4, 2026Yuxin Liao, Le Wu, Min Hou +3LLM Agent Memory Persistent Memory for Language Models
Hefei University of Technology
LycheeMemory V2: Efficient Long-Term Memory for LLM Agents via Semantic Segment-Level Consolidation
Aug 13, 2026Dongfang Li, Zixuan Liu, Junmai Wang +6LLM Agent Memory Persistent Memory for Language Models
Harbin Institute of Technology, Shenzhen

Sep 27, 2026J/K move · Enter open · S save

ActiveMem: Dynamic Latent Memory Trees for Long-Horizon Agents

Song-Li Wu, Jingyi Wang, Zhaocheng Du, Weinan Gan

Tsinghua University · Huawei Noah’s Ark Lab

LLM Agent Memory Memory-Augmented Language Model Agents Agent Memory Management Long-Horizon LLM Agents

Open paper AskPDF arXiv

Large Language Model (LLM) agents increasingly rely on external memory to support long-horizon reasoning and decision making. Existing memory systems typically retrieve historical trajectories or summaries as independent context fragments, overlooking the procedural dependencies underlying multi-step execution. As memory scales, such flat retrieval introduces context fragmentation and cross-task interference, leading to structurally inconsistent reasoning trajectories. We propose ActiveMem, a hierarchical memory framework that recursively organizes agent experiences into dependency-aware latent execution trees. ActiveMem abstracts trajectories into reusable subtask nodes while explicitly preserving execution transitions, enabling coherent reasoning-path retrieval conditioned on the current execution state. To support continual adaptation, ActiveMem further learns dynamic memory expansion, retrieval, and pruning policies through reinforcement learning. Experiments across various agent benchmarks demonstrate that ActiveMem consistently improves task completion, reasoning stability, and memory efficiency over existing memory-based agents. Moreover, ActiveMem enables compact open-weight models to achieve competitive performance with substantially larger proprietary systems.

Abstract

LLM Agent Memory

Explore similar work

ActiveMem: Dynamic Latent Memory Trees for Long-Horizon Agents

LeanMem: Simple and Efficient Long-Term Memory for LLM Agents

LycheeMemory V2: Efficient Long-Term Memory for LLM Agents via Semantic Segment-Level Consolidation