AgriGPT: Generative AI for Agriculture

• AgriGPT is a suite of generative AI systems tailored for agriculture, coupling language models with sensor data and multimodal analytics for precision decision-making.
• These systems employ diverse methodologies such as graph neural networks, retrieval-augmented generation, and vision-language models to support crop recommendation, weather forecasting, and pest detection.
• Variants like AgriGPT-VL and AgriGPT-Omni illustrate advances in vision-language integration and multilingual capabilities, broadening agricultural advisory and cross-linguistic query resolution.

Searching arXiv for the cited AgriGPT-related papers to ground the article in current records. {"query":"AgriGPT (Potamitis, 2023, Thapa et al., 8 Jun 2026, Yang et al., 5 Oct 2025, Yang et al., 11 Dec 2025, Awais et al., 2024, Didwania et al., 2024, Zhao et al., 2023, Silva et al., 2023, Yang et al., 2024, Zaremehrjerdi et al., 25 May 2025)","max_results":10} {"query":"AgriGPT agriculture GPT multimodal agricultural assistant", "max_results": 10} AgriGPT denotes a class of agriculture-specialized generative AI systems that couple LLMs or multimodal LLMs with agricultural data, interfaces, and decision workflows. In the arXiv literature, the label has been applied to at least four distinct but related configurations: a speech-driven analytics layer over precision-agriculture sensor databases, a mobile crop-recommendation and agricultural question-answering system built on spatio-temporal graph neural networks and hybrid retrieval augmentation, a domain-specialized agricultural vision-LLM, and a unified speech-vision-text framework for multilingual agricultural intelligence (Potamitis, 2023, Thapa et al., 8 Jun 2026, Yang et al., 5 Oct 2025, Yang et al., 11 Dec 2025). Related systems such as AgroGPT, AgriLLM, and ChatAgri occupy the same design space, emphasizing expert tuning, farmer-query resolution, and cross-linguistic agricultural text classification (Awais et al., 2024, Didwania et al., 2024, Zhao et al., 2023). This suggests that AgriGPT is best understood as a research family rather than a single canonical architecture.

1. Terminological scope and research lineage

The term emerged in work on precision-agriculture data analytics, where ChatGPT was positioned as an interface to heterogeneous field data generated by a dispersed network of automated insect traps and sensor probes. In that setting, the central proposition was that speech recognition, natural-language querying, Python/Pandas execution, visualization, and speech synthesis could be integrated into a single analytics loop for policymakers and agronomists (Potamitis, 2023). Subsequent uses of the name broadened the concept substantially. One line of work defined an AgriGPT system around 30-day weather forecasting, crop suitability scoring, and a Retrieval-Augmented Generation chatbot deployed through a mobile application (Thapa et al., 8 Jun 2026). Another line used AgriGPT to denote a domain-specialized vision-language suite centered on agricultural corpora, curriculum alignment, and GRPO refinement (Yang et al., 5 Oct 2025). A further extension, AgriGPT-Omni, generalized the idea to multilingual speech, vision, and text in a unified omni-model (Yang et al., 11 Dec 2025).

Related naming conventions are important. AgroGPT is an efficient agricultural vision-LLM trained by expert tuning on AgroInstruct, a synthetic multimodal conversation set derived from vision-only agricultural datasets (Awais et al., 2024). AgriLLM focuses on transformer-based automation of farmer helpline queries using approximately 4 million Tamil Nadu call-center records (Didwania et al., 2024). ChatAgri studies zero-shot and prompt-based agricultural text classification across English, French, and Chinese (Zhao et al., 2023). These systems are not identical, but collectively they define the operational meaning of AgriGPT in current research: domain-grounded language or multimodal intelligence for agricultural analytics, advisory, and reasoning.

System family	Primary setting	Distinguishing components
AgriGPT	Precision-agriculture analytics	STT, semantic parsing, Python/Pandas, visualization, TTS
AgriGPT	Crop recommendation and QA	STGCN, hybrid retrieval, mobile deployment
AgriGPT-VL	Agricultural vision-language understanding	Agri-3M-VL, curriculum alignment, GRPO
AgriGPT-Omni	Multilingual speech-vision-text reasoning	Unified omni-model, six languages, tri-modal benchmark
Related systems	Text or multimodal advisory	AgroInstruct, KCC fine-tuning, cross-lingual prompting

2. Textual advisory systems and agricultural query answering

Text-centric AgriGPT-like systems span prompt-only pipelines, fully fine-tuned Seq2Seq models, and retrieval-augmented exam or advisory assistants. ChatAgri frames agricultural text classification as a prompt-construction, ChatGPT Q&A, and answer-alignment problem. It evaluates manually defined prompts, ChatGPT-triggered prompts, zero-shot similarity prompts, and chain-of-thought prompts on sentiment analysis and topic classification tasks across English, Chinese, and French. On the reported zero-shot results, ChatAgri reaches 0.978 accuracy and 0.978 weighted-$F_1$ on Natural-Haz-Type, 0.863 and 0.856 on Agri-News-Chinese, 0.794 and 0.789 on PestObserver-FR, and 0.798 and 0.793 on Amazon-Food; the GPT-4 variant improves cross-lingual robustness by approximately 2–3 points over GPT-3.5 (Zhao et al., 2023). The system is notable because it treats agricultural language understanding as an immediate deployment problem under low-label, cross-linguistic conditions rather than as a conventional supervised fine-tuning task.

AgriLLM represents a different design choice: large-scale supervised adaptation on real farmer helpline logs. It uses Kisan Call Center transcript logs collected by the Indian Ministry of Agriculture & Farmers Welfare from 2006–2023, with a 4 million-record Tamil Nadu subset stratified by