GridProbe

GridProbe is an evidence-seeking AI diagnostic agent for physical infrastructure. Instead of sending telemetry to an LLM and asking for a guess, GridProbe actively decides what evidence it needs next, calls deterministic analysis and retrieval tools, records every observation in a traceable Evidence Ledger, and only produces a structured diagnosis when the evidence is sufficient. The current prototype focuses on electrical panels and can investigate overload, phase imbalance, connection-related heating, and insufficient-evidence cases. It combines Meta Muse Spark 1.3 reasoning with deterministic Python tools, structured evidence tracking, explicit abstention, and a multimodal architecture using SAM 3.1 and Muse visual observation. When the evidence cannot establish a cause, GridProbe returns NEEDS_MORE_EVIDENCE instead of inventing a diagnosis. The system is designed as decision-support software for qualified maintenance teams and as a hardware-ready foundation for future industrial telemetry integrations.