HORMA: Hierarchical Memory System Improves LLM Agent Performance on Long-Horizon Tasks
Researchers have introduced HORMA, a hierarchical memory architecture for large language model (LLM) agents designed to handle long-horizon tasks more efficiently. Current LLM agents struggle with statelessness, forcing all relevant information into ever-growing input contexts, which degrades reasoning and increases cost. HORMA addresses this by organizing experience into a file-system-like structure and using reinforcement learning-trained navigation, reducing token usage to as little as 22% of baseline in long conversation tasks.
A team of researchers has proposed HORMA (Hierarchical Organize-and-Retrieve Memory Agent), a new working memory framework for LLM-based agents that tackles a core limitation: the inability to efficiently manage information across long, multi-step tasks. Unlike existing methods that rely on lossy compression or similarity-based retrieval, HORMA organizes agent experiences into a hierarchical, file-system-like structure where summarized entities are linked to their underlying raw trajectories. The system separates memory into two stages—structured construction and navigation-based retrieval—with the construction module distinguishing between failures caused by missing information versus those caused by misleading or overloaded context. A lightweight reinforcement learning agent then traverses the hierarchy to retrieve only the minimal context necessary for a given task. Evaluated on ALFWorld, LoCoMo, and LongMemEval benchmarks, HORMA demonstrated improved task performance under constrained context budgets while using at most 22.17% of the token count required by baseline approaches. The authors report that the system generalizes effectively to unseen tasks and consistently achieves better efficiency-performance trade-offs compared to prior methods.
What's missing
The paper does not report results from independent replication or external peer review, as it is a preprint. Key limitations not discussed in the abstract include computational overhead of the RL-trained navigation agent itself, and how performance scales with very large or highly dynamic task environments. The scope of the benchmark tasks (ALFWorld, LoCoMo, LongMemEval) may not fully represent real-world agentic deployment conditions.
What different sources said
- arXiv cs.AICenter
Organize then Retrieve: Hierarchical Memory Navigation for Efficient Agents
Related
Gut Bacteria Enzyme Found to Break Down Heat-Processed Food Compounds, Producing Novel Biogenic Amines
Researchers have discovered that an enzyme in common gut bacteria can degrade N-epsilon-carboxymethyllysine (CML), a compound formed during thermal food processing, producing previously unknown biogenic amines. The enzyme, ornithine decarboxylase SpeC from enterobacteria, acts on CML and related modified lysine derivatives through a low-level 'underground' catalytic activity. This finding suggests a previously unrecognized communication axis between thermally processed dietary compounds and gut microbial physiology, with potential implications for host health.
Full-Length Gene Sequencing Reveals Two Distinct Bacterial Communities in Black-Legged Ticks Expanding Into Canada
Researchers used Oxford Nanopore full-length 16S rRNA gene sequencing to characterize the microbiome of Ixodes scapularis black-legged ticks collected in Nova Scotia, Canada, distinguishing between tick-adapted bacteria and environmentally acquired bacteria. The study comes as I. scapularis — the primary vector of Lyme disease — is rapidly expanding northward into Canada due to climate change. The findings suggest that environmentally derived bacteria in tick microbiomes are not mere contamination, which has implications for how tick microbiome data is collected and interpreted across surveillance studies.
Study Identifies Metabolic Link Between Cell Envelope Stress and Biofilm Formation in Bacteria
Researchers have discovered that the metabolite acetyl-CoA directly inhibits enzymes that degrade the bacterial signaling molecule c-di-GMP, connecting cell envelope biosynthesis stress to biofilm formation in Pseudomonas aeruginosa. The study found that sub-inhibitory concentrations of antibiotics targeting early peptidoglycan biosynthesis — but not other antibiotic classes — elevate c-di-GMP levels by reducing phosphodiesterase activity, with acetyl-CoA competing for the enzyme active site. Because the relevant enzyme domain is broadly conserved across bacterial species, this checkpoint mechanism may be widespread and could have implications for understanding antibiotic-induced biofilm responses.