Eluna: An Agentic LLM System for Automating Warehouse Operations with Reasoning and Task Execution

2026-07-09

AI Summary

This work addresses the challenge of executing complex standard operating procedures (SOPs) in warehouse operations, where existing large language model agents struggle to balance timeliness and reliability due to context overload and insufficient compliance mechanisms. The authors propose a graph-guided multi-agent system that encodes SOPs as directed acyclic graphs and employs progressive information disclosure to orchestrate specialized sub-agents capable of persistent code execution and real-time data access. A novel asymmetric scenario distillation mechanism enables a lightweight student model to internalize the teacher’s error-correction experience, achieving high accuracy and low latency without relying on memory during inference. Evaluated across 13 tasks and two production deployments, the approach matches or exceeds the teacher model’s performance and significantly outperforms larger baselines, attaining 94% agreement with human experts on work order processing.

Abstract

Warehouse operations are governed by Standard Operating Procedures (SOPs) that encode complex, multi-system decision logic, which must be executed reliably under strict time constraints, yet LLM agents lack mechanisms to enforce procedural compliance and degrade under the context overload full SOP specifications introduce. We present Eluna, a production-deployed agentic system for reliable SOP execution. Eluna is a graph-guided, multi-agent framework that encodes SOPs as directed acyclic graphs with progressive disclosure and delegates independent tasks to parallel sub-agents, each with persistent code execution and live data access. To meet production latency and accuracy needs, we use asymmetric episodic distillation where a strong teacher is improved through episodic error memories, then a smaller student is fine-tuned on the corrected trajectories with memory stripped, internalizing corrections without inference-time overhead. On a 13-task benchmark and two production applications, our fine-tuned models match or exceed their teacher, beat all larger off-the-shelf baselines, and reach 94% expert agreement on the ticket processing application.

Problem

• Standard Operating Procedures
• LLM agents
• procedural compliance
• context overload
• warehouse operations

Innovation

• graph-guided multi-agent
• SOP execution
• asymmetric episodic distillation
• progressive disclosure
• persistent code execution