Transformer APM workbench

AgenticGrid Pro

A local-first asset-performance-management workbench for power transformers. Evidence goes in, agents draft the reasoning, and a named engineer signs off before anything becomes a recommendation.

What it is

One workbench for the transformer decision

AgenticGrid Pro installs on a Windows workstation inside your perimeter and holds the whole chain of reasoning about a transformer in one place.

Transformer condition evidence arrives in fragments: a laboratory dissolved-gas report as a PDF, a partial-discharge test set export, an SFRA sweep against a baseline that lives in a different folder, thermal survey imagery, a nameplate record, and a decade of work orders. Assembling those fragments into something a reviewer can act on is most of the work, and it is done by hand.

AgenticGrid Pro is the workbench for that work. It parses each source, keeps its provenance, aligns the signals against one asset record, screens them for gaps and unit errors, and then lets an agentic AI workflow draft a condition assessment with the evidence chain attached. Nothing it drafts leaves the workbench until an engineer has put their name on it.

It is decision support, not automation. AgenticGrid Pro performs no switching, sets no protection, and holds no operational authority.

Who it is for

Built for the people who own the unit

The workbench assumes an engineering reader, not a dashboard viewer.

Transformer and diagnostics engineers
The reviewers who read a Duval placement, a PRPD pattern family and an SFRA deviation band, and who are asked to justify every recommendation.
Substation and maintenance planners
Teams turning condition findings into outage windows, task lists and CMMS work orders, with the evidence attached to each line.
Asset-management and reliability leads
Owners of fleet prioritisation who need driver attribution behind a health score, not an unexplained number.
Security and procurement reviewers
The people who have to approve where the software runs, what leaves the network, and who signs the outputs.

The story

From raw evidence to a signed work package

The product has one shape, and every screen serves it. Evidence is taken in with its source, timestamp and owner recorded. Signals are correlated per asset and screened by a quality gate before any reasoning starts, so rejected inputs are logged with the reason they failed rather than quietly dropped.

Agents then draft a condition assessment that cites each piece of evidence it relied on, states a confidence band, and flags the signal that disagrees instead of smoothing it over. A named engineer approves, edits, rejects or escalates that draft. Only after that does the workbench assemble a maintenance work package and an evidence-pack report — each line traceable back to the records it was built from.

The result is not a prediction you have to take on faith. It is a document your own reviewers can audit, and that your regulator, insurer or board can follow backwards.

The human gate

The AI drafts. A named engineer decides.

Stage four of the workflow is mandatory and cannot be configured away. Every AI-drafted assessment carries a persistent draft marker until a qualified engineer records a decision, and that decision — the choice, the reviewer, the comment, the timestamp — is written to an append-only audit log.

This is why the engineering authority in AgenticGrid Pro is deterministic and local. Gas ratios, trend rates, deviation bands and thermal aging estimates are computed in code you can inspect the inputs and assumptions of. The language model contributes drafting, correlation narrative and retrieval — not the number a decision rests on.

Why now

The fleet is old and the replacement is slow

The U.S. Department of Energy estimates that 90 percent of all electricity consumed in the United States passes through a large power transformer at some point between generation and end user, and in 2022 DOE estimated 4,900 to 6,799 of those units in service.

Those units are not young. Large power transformers are typically considered to have a design lifetime on the order of 40 years, and a 2014 DOE report — restated in DOE's 2024 Report to Congress — estimated the average age of large power transformers in the North American grid at 38 to 40 years, meaning a substantial fraction is at or past design lifetime.

Replacing one is no longer a quick answer either: the same DOE report to Congress states that lead times for acquiring a large power transformer have become exceptionally long, with 36-month lead times commonly quoted and maximum lead times reaching as much as 60 months. When the spare is years away, the quality of the decision you make about the unit you have is what matters.

The product family

One of GridAPM's two products

GridAPM builds two products. AgenticGrid Pro is the power-transformer asset-performance-management workbench described on these pages. ProtectionAI is GridAPM's protective-relay testing software with an agentic AI copilot, aimed at relay test engineers rather than transformer asset teams.

They are separate applications with 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.

Put your own evidence through the workflow

A bounded evaluation starts with the records you already have and ends with an evidence pack your reviewers can inspect line by line. Your units, your engineers at the gate.

Type to search research, platform pages, and tools.