A hierarchical memory architecture overcomes context limits in long-horizon multi-agent computational modeling
Abstract
Large language models (LLMs) demonstrate remarkable reasoning capabilities, yet their stateless architecture fundamentally limits deployment in long-horizon research workflows requiring multi-session continuity and quantitative rigor. Here we present Ensemble QSP, a multi-agent framework featuring a three-layer hierarchical memory architecture that bounds injected context (median 301 tokens, max 4,050) by capping state categories and evicting completed work. This enables continuous autonomous operation without context degradation. The system orchestrates five specialist worker agents under domain-expert principal investigators (PIs), enforcing physical constraints through physics-based checklists and structured domain knowledge. Comprehensive benchmarking demonstrates autonomous pharmacokinetic-pharmacodynamic (PKPD) model selection, improved parameter recovery relative to single-agent baselines, and robust interpretation of linguistically diverse prompts. Replication with open-weight models (DeepSeek-V4-Flash/Pro, Llama 3.1 70B) confirmed these architectural conclusions across PKPD modeling, literature synthesis, and PBPK model implementation, proving the framework is independent of proprietary LLMs. Feature-level ablations show that memory, retrieval, and PI oversight address distinct scientific failure modes, though underlying LLM capability remains consequential for stringent physical-consistency checks. The architecture is structurally agnostic to computational biology; adding a new scientific domain requires only a new PI-agent configuration.
Related
- A context-aware temporal modeling through unified multi-scale temporal encoding and hierarchical sequence learning for single-channel EEG sleep staging
- Scaffold-Constrained Generative Design Overcomes Failure Modes in Computational Protein Engineering
- Chronotherapy leveraging dietary context overcomes circadian rhythm influence on cancer cell response to PPAR antagonists
- HSCM-Lane: A ResNet-based encoder-decoder architecture with multi-level shifted-window context modeling for pixel-wise lane segmentation
- Revealing multiscale architecture of the intact human spine using hierarchical phase contrast tomography
BibTeX
@misc{tewari2026hierarchical,
title = {{A hierarchical memory architecture overcomes context limits in long-horizon multi-agent computational modeling}},
author = {Shivendra G. Tewari and Holly Kimko},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2607.07666},
eprint = {2607.07666},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2607.07666},
}
RIS
TY - UNPB
TI - A hierarchical memory architecture overcomes context limits in long-horizon multi-agent computational modeling
AU - Shivendra G. Tewari
AU - Holly Kimko
PY - 2026
DA - 2026/09/17
PB - arXiv
DO - 10.48550/arxiv.2607.07666
UR - https://arxiv.org/abs/2607.07666
AB - Large language models (LLMs) demonstrate remarkable reasoning capabilities, yet their stateless architecture fundamentally limits deployment in long-horizon research workflows requiring multi-session continuity and quantitative rigor. Here we present Ensemble QSP, a multi-agent framework featuring a three-layer hierarchical memory architecture that bounds injected context (median 301 tokens, max 4,050) by capping state categories and evicting completed work. This enables continuous autonomous operation without context degradation. The system orchestrates five specialist worker agents under domain-expert principal investigators (PIs), enforcing physical constraints through physics-based checklists and structured domain knowledge. Comprehensive benchmarking demonstrates autonomous pharmacokinetic-pharmacodynamic (PKPD) model selection, improved parameter recovery relative to single-agent baselines, and robust interpretation of linguistically diverse prompts. Replication with open-weight models (DeepSeek-V4-Flash/Pro, Llama 3.1 70B) confirmed these architectural conclusions across PKPD modeling, literature synthesis, and PBPK model implementation, proving the framework is independent of proprietary LLMs. Feature-level ablations show that memory, retrieval, and PI oversight address distinct scientific failure modes, though underlying LLM capability remains consequential for stringent physical-consistency checks. The architecture is structurally agnostic to computational biology; adding a new scientific domain requires only a new PI-agent configuration.
ER -