VetClaw Separates Agent Interaction From Workflow Orchestration by Pairing OpenClaw With LangGraph on an Edge Device

arXiv 2607.26042·low signalXBlueskyLinkedInCopy link

VetClaw (arXiv 2607.26042, 2026-07-28) is an edge-cloud multimodal agentic system for early veterinary disease screening, and its architectural split is the transferable part: OpenClaw handles scheduling, tool access, user interaction, and notifications on the edge device, while LangGraph owns the stateful screening workflow including input validation, image transmission, model invocation, safety checks, conditional routing, failure handling, and structured logging. A camera module captures images that, with optional symptom descriptions, go to a server-hosted VLM for zero-shot classification. Results show image-only VLM prediction remains limited while symptom-guided and multimodal inputs improve zero-shot performance — the system's value comes from deterministic safety rules and escalation of uncertain cases, not from raw classifier accuracy.