Reproduce “ALAS: A Stateful Multi-LLM Agent Framework for Disruption-Aware Planning” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Edward Y. Chang, Longling Geng

The ALAS framework enables Large Language Models to handle complex, disruption-aware planning tasks by introducing a stateful multi-agent architecture. This system leverages persistent memory and layered orchestration to overcome limitations like context erosion and next-token myopia, demonstrating improved performance across diverse domains such as urban ride sharing, family reunion planning, and job shop scheduling.

Key insights

  • Decomposing complex planning problems into specialized agents, each with persistent memory, helps LLMs manage context and long-term consequences.
  • A three-layer architecture (Template, Factory, Runtime) enables systematic workflow generation and orchestration, addressing next-token myopia and lack of self-verification.
  • Integrating disruption-aware compensation mechanisms at runtime allows localized recovery and maintains plan consistency without full replanning.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.