AgenticGrid Pro questions

It decides nothing operational. AgenticGrid Pro assembles transformer evidence, gates its quality, computes deterministic diagnostics, and drafts a condition assessment with the evidence chain attached. Whether that assessment becomes a recommendation is decided by a named engineer in your organisation. It performs no switching, sets no protection, and holds no final diagnostic or operational authority.

No. The AI drafts; a named engineer signs off. Stage four of the workflow is mandatory: a drafted assessment carries a persistent draft marker until a qualified engineer approves, edits, rejects or escalates it, and that decision — with the reviewer's identity, comment and timestamp — is written to an append-only audit log. Only an approved assessment can become a work package.

Only the generative calls you configure, to the provider you configure. AgenticGrid Pro makes outbound calls solely for generative features an operator requests, using your own OpenAI or Anthropic credential and carrying the context needed for that operation. The operational database, deterministic analysis, review workflow and sign-off record stay on the workstation. Product telemetry is off, and there is no inbound connection into the OT network at all.

The deterministic workbench can. Intake, quality gating, diagnostic computation, review and engineer sign-off, work-package assembly and report export all run locally, and trial and licence checks are validated locally rather than against a GridAPM service. What requires network access is the AI-drafted narrative and provider-based semantic retrieval; configuring retrieval to use deterministic local feature hashing removes that dependency at the cost of recall. On a fully air-gapped machine you get the deterministic workbench without the generative drafting.

Less than teams usually assume. A useful start is an asset list with identifiers and nameplate context, plus one primary stream — most often laboratory DGA history — and one or two supporting streams such as PRPD, SFRA, thermal surveys, loading extracts, inspections or maintenance history. What matters more than volume is that records are dated, unit-consistent and attributable to a known asset.

Yes, and that is the expected case. AgenticGrid Pro reads laboratory reports as PDF, CSV or XLSX from whichever laboratory you use, and preserves the laboratory identity, sampling date, units and detection limits as part of the record. The quality gate reconciles duplicate reports and checks unit and sampling-date consistency before any value enters a trend. Exports from online gas and moisture monitors are supported alongside the laboratory record, cross-checked against the most recent lab sample.

No. This is context, not certification. GridAPM holds no certification, accreditation or endorsement from IEEE, IEC, NIST or any other body, and none of them has reviewed or approved the product. Naming IEEE C57.104, IEC 60599, IEEE C57.91 or the NIST AI Risk Management Framework describes the frameworks our evidence model and AI governance posture are organised around. Interpretation remains subject to qualified engineering review and your own procedures.

Not as a live connection. AgenticGrid Pro reads exported files and approved extracts: historian CSV extracts, CMMS or EAM extracts your team produces, monitor exports. It never connects to relays, RTUs, protection, control or SCADA systems. On the way out, work packages export as CSV for CMMS or EAM import, so no write access to your system of record is required.

It is yours. The evidence pack is the deliverable: condition evidence, charts, the AI rationale, reviewer comments and the complete append-only sign-off trail, assembled so a third party can trace every conclusion back to a source record. It exports as PDF from the workstation. At the end of an engagement, source evidence is returned or deleted to the schedule agreed at scoping, while the pack itself is handed over as the auditable record of what was decided.

A bounded evaluation on your own transformers rather than an open-ended trial. Evidence intake and quality gating come first, then agent-drafted analyses reviewed by your engineers in cycles, then an evidence pack and a value summary scoped to the criteria you set at the start. Your side provides named reviewers, approved historical records, the decision workflow to evaluate and a workstation inside your perimeter. Outcomes are framed as hypotheses to test on your fleet, not as results we can promise in advance.

They are two separate products from one company. AgenticGrid Pro is the power-transformer asset-performance-management workbench. ProtectionAI is GridAPM's protective-relay testing software with an agentic AI copilot, built for relay test engineers. Separate applications, separate licences, and neither requires the other. What they share is the engineering philosophy: local-first deployment, deterministic calculation, AI that drafts and cites rather than decides, and a named human at the gate.

No, and the design goes out of its way to prevent that reading. Every score is presented with its driver attribution, trend movement, the weighting assumptions applied, and an explicit note on uncertainty and missing evidence. A reviewer should be able to say which evidence moved a unit up the priority list and what would change if a missing test arrived — from the workbench alone.

The draft says so. When one stream suggests a developing fault and another does not support it, the agent flags the disagreement explicitly rather than resolving it silently toward the more alarming reading, and the draft carries a confidence band. Surfacing contradictions is treated as a feature of the output, because a reviewer who is told where the evidence is thin can act on it.

No. Demonstration content across the site uses a labeled synthetic specimen asset, TX-47 · 230/69 kV, with synthetic values. No customer data, customer name or customer result appears anywhere on this site, and GridAPM publishes no performance benchmark of its own.