Agri-SAGE: Simulation-Grounded Multi-Agent LLM for Context-Aware Agricultural Advisory Generation
06:00 · July 2, 2026 · arXiv cs.AI RSS
Agricultural advisory systems face a fundamental tension: static agronomic guidelines offer consistent, evidence-based recommendations, yet remain blind to in-season variability and dynamic uncertainties. Recent advisory systems powered by LLMs are liable for a different risk of generating recommendations that are agronomically credible but physiologically unconvincing. Agri-SAGE is a closed-loop framework designed to resolve the above two limitations by integrating retrieval-grounded multi-agent LLM reasoning with APSIM-based biophysical simulation, to generate and validate agronomic advisories. To assess this framework, we evaluate three reasoning approaches, namely Plan-and-Solve, Tree of Thoughts, and Reflexion, over a 10-year retrospective analysis. All three significantly outperform static PoP (Package-of-Practice) baselines, with Tree of Thoughts achieving impressive peak yields. At the same time, Reflexion achieves comparable agronomic outcomes at substantially lower computational cost by leveraging cross-seasonal episodic memory.
Summary
Agri-SAGE tackles the core shortcomings of conventional agricultural advisory systems by coupling retrieval-augmented multi-agent LLM reasoning with the APSIM biophysical crop simulator. Static Packages of Practice supply evidence-based guidelines yet cannot respond to intra-season weather shifts, pest pressure, or local soil differences, while standalone LLM systems risk producing recommendations that sound agronomically plausible but fail basic physiological checks. The framework closes this gap through a retrieval agent that grounds prompts in regional agronomic documents, a generation agent that explores management sequences, and a verification agent that routes every candidate plan through APSIM for daily-resolution simulation of crop growth, water balance, and nutrient dynamics.
Three iterative reasoning strategies—Plan-and-Solve, Tree of Thoughts, and Reflexion—were tested in a ten-year retrospective study on maize in the Mandya region. All three markedly exceeded the yields obtained from fixed PoP baselines. Tree of Thoughts produced the largest yield gains by maintaining an explicit search tree of candidate actions and backtracking on poor APSIM outcomes. Reflexion reached comparable agronomic performance at substantially lower compute cost by storing cross-seasonal episodic memory that lets the agent reuse successful prior trajectories rather than re-exploring the full decision space each season.
The closed-loop design ensures that every advisory remains both context-aware and physiologically feasible before it is presented to the farmer. By feeding APSIM observations back into the LLM agents, Agri-SAGE converts what would otherwise be open-ended text generation into an iterative optimization process anchored in process-based crop modeling.
Why it matters
This research is highly relevant to the Dutch AI market given the Netherlands' status as a global leader in agritech and precision agriculture. The integration of multi-agent LLMs with biophysical simulations offers actionable, advanced methodologies for Dutch researchers and enterprises looking to optimize crop yields and agricultural sustainability.
