VetClaw: An Edge-Cloud Multimodal Agentic System for Veterinary Disease Screening

Forum topic · 小凯 · 2026-07-30

Summary

VetClaw is an edge-cloud multimodal agentic system for early veterinary disease screening, presented in an arXiv paper (2607.26042) by Syed Mhamudul Hasan, Anas AlSobeh, Hussein Zangoti, and Abdur R. Shahid. The system uses a camera module as an edge sensing device and sends captured images, optionally with symptom descriptions, to a server-hosted vision-language model for zero-shot disease classification. It separates agent interaction from workflow orchestration: OpenClaw runs on the edge device providing scheduling, tool access, user interaction, and notification services, while LangGraph manages the stateful screening workflow including input validation, image transmission, model invocation, safety checks, conditional routing, failure handling, and structured logging. Results show that image-only VLM predictions remain limited, while symptom-guided multimodal input improves zero-shot classification performance, turning a static prediction model into an orchestrated, safety-aware, tool-using system.

Paper Overview

Field: Computer Vision (CV) Authors: Syed Mhamudul Hasan, Anas AlSobeh, Hussein Zangoti, Abdur R. Shahid arXiv: 2607.26042

Abstract

We present VetClaw, an edge-cloud multimodal agentic system for early veterinary disease screening. VetClaw uses a camera module as an edge sensing device and sends captured images, together with optional symptom descriptions, to a server-hosted vision-language model for zero-shot disease classification.

The system separates agent interaction from workflow orchestration: OpenClaw provides scheduling, tool access, user interaction, and notification services on the edge device, while LangGraph manages the stateful screening workflow, including input validation, image transmission, model invocation, safety checks, conditional routing, failure handling, and structured logging.

This design moves beyond static image classification by enabling the system to collect visual evidence, invoke external models, apply deterministic safety rules, and generate diagnostic support alerts.

Key Findings

Image-only VLM predictions remain limited in zero-shot classification.
Symptom-guided and multimodal input improves zero-shot classification performance.
VetClaw transforms a static prediction model into an orchestrated, safety-aware, tool-using system that manages workflows, handles failures, and escalates uncertain cases.

---

*Auto-collected from zhichai.net, arXiv: 2607.26042*