The questions engineersactually ask

Only within stated limits. ProtectionAI ships a bounded OMICRON CM Engine adapter — device discovery, lock and unlock, static voltage and current phasors, output control and software-paced ramps, driven through your own licensed CM Engine installation — and physical-CMC hardware-in-the-loop qualification is still pending, so the built-in simulator remains the only fully qualified path. A generic SCPI source driver speaks to a programmable AC source over TCP but has no binary sensing, so it cannot detect a trip. The Doble, Megger, ISA/Altanova and EuroSMC drivers are documented stubs that refuse to connect. If your requirement is a qualified hardware workflow, this version cannot meet it yet.

The shipping build can bind a selected interface through SharpPcap/Npcap for GOOSE and Sampled Values capture/injection, while retaining deterministic loopback for rehearsal, and it never silently falls back between the two. This is not automatically substation-qualified: Npcap licensing and permissions, network isolation, live NIC/IED validation and an OT change process are required. Capture timestamps identify whether they came from the application, software pcap or a hardware-capable path.

No. The copilot is optional and it requires an API key you supply — Anthropic, OpenAI, xAI/Grok or Google/Gemini. GridAPM does not sell or resell model capacity. If you never configure a key, the copilot panel stays inert and the rest of the application works exactly as it otherwise would. Everything the copilot can do — running tests, editing the working settings model, driving output phasors, generating and importing templates, searching your ingested manuals, drafting reports — is available to you manually. The copilot removes typing; it does not hold capabilities hostage.

Only if you enable the copilot, and then only to the provider you configured. With no API key set, nothing is sent anywhere and the whole workflow stays local. With a key set, ProtectionAI calls your chosen provider directly from your workstation, sending the material needed for the step you asked for: your prompt, the relevant settings, retrieved passages from manuals you ingested, results being interpreted, and attachments you add. There is no GridAPM endpoint in that path, so your provider's terms are the ones that apply. Manual search over your ingested PDFs runs locally.

No, and no software is — there is no NERC certification for a testing application. Compliance under a standard such as PRC-005 is a property of your maintenance programme and your evidence, assessed by your regional entity. What ProtectionAI provides is the record layer plus the tracking layer: test sessions filed against a specific relay asset with the settings version, operator, test set, measured and expected values, per-point verdicts and attachments, all retrievable by asset; and on top of that, per-asset maintenance programmes, NERC interval defaults you can edit per component type and monitoring level, a Compliant, Due Soon and Overdue status per asset, a KPI dashboard, and exportable evidence packages with a manifest. The remaining gap is multi-user approval routing — sending a completed record through a reviewer and an approver — which is a roadmap item.

The application, without a key, for seven days. Because the licensed product also runs its tests against the built-in simulator today, the trial is a fair representation of the working software rather than a cut-down demonstration. Import your own XRIO or RIO settings, build the plans you would really run, execute them, review the measured-against-expected results and export a report on your own letterhead. Then decide.

Email sales@gridapm.com. Licences are RSA-signed and bound per computer, so a machine replacement or a reimage needs a reissued key rather than a self-service transfer. Tell us the machine being retired and the one replacing it. If machine churn is a normal part of your environment, raise it during procurement so the licence terms cover it explicitly.

Yes, fully. No internet connection is required to run tests or to activate a licence — activation is offline and validated locally. The installer is self-contained, so installation does not fetch a runtime either. The single exception is the optional copilot, which needs an outbound path to the AI provider you configured. An isolated machine is a supported and complete installation, minus the copilot.

Three things: a virtual test set that accepts the same channel, ramp, sequence and playback commands a real driver would; a configurable simulated relay that responds to them, including realistic contact timing; and a power-system network model you draw in a single-line editor, solved for short-circuit conditions and played back closed-loop, with the breaker opening at the next current zero. So a 50/51 sweep really searches for operate times, a distance sweep really searches for a zone boundary, a differential search really finds an operate point and a sequence really checks the binary transitions you declared. What it cannot do is reveal a wiring error, a CT polarity mistake or a firmware quirk in your actual relay — it is a model of a relay, not your relay.

They are hardware-anchored suites: each is the control software for its own vendor's instruments, with decades of element test modules, large relay-specific template libraries and mature compliance databases attached. On breadth of modules, template libraries, system-based testing, real-network 61850, synchronised end-to-end testing and hardware control, they are ahead — obviously, in the last case. ProtectionAI starts from a different point: a hardware-neutral platform where the intelligence and record layers come first, with one workflow across execution, settings management and documentation, open JSON templates and offline-first licensing. We are not going to claim we replace twenty years of vendor modules; we publish an honest capability matrix instead, and you should read it before you compare anything.

One thing we can defend as a dated research finding: in GridAPM research across the published OMICRON, Doble, Megger, ISA and Altanova, and SMC and EuroSMC protection portfolios as of July 2026, we found no AI-branded capability in their protection-testing product material. ProtectionAI ships a tool-using agent with retrieval over your own relay manuals, a central isolation-confirmation gate for tools that can energize outputs, and deterministic code that retains every verdict. That is our reading of public vendor sources on that date, not a permanent claim about anybody's roadmap. The other genuine difference is the simulator-first workflow: you can rehearse a complete test programme before you lift a lead.

In: OMICRON XRIO and RIO parameter files into a normalised settings model, native SEL ASCII, DIGSI 5 XML and generic name/value CSV/TXT settings files chosen by content sniff, IEC 61850 SCL (SCD, ICD, CID), COMTRADE (IEEE C37.111) records, JSON test templates and plans, and PDF relay manuals for local retrieval and as asset attachments. Out: COMTRADE for simulator captures, CSV results, JSON templates and plans, SCL under a preserve-verbatim-or-refuse writer, branded Word, Excel and PDF reports carrying your own logo, plus file-based hand-off — an .ics calendar and CMMS-import CSV. Importers for GE, ABB PCM600, MiCOM S1, Beckwith and Basler settings are on the roadmap, and there is no live SCADA, EAM or historian API integration today.

No. ProtectionAI is a Windows desktop application built on .NET 9 WPF, supported on Windows 10 and Windows 11 x64. There is no macOS or Linux build, and none is planned. It runs acceptably in a Windows virtual machine, but you should evaluate that yourself during the trial rather than assume it.

AgenticGrid Pro is GridAPM's other product: asset performance management software for power transformers. It is a different product for a different job and neither depends on the other. If your interest is relay testing, ProtectionAI is the one to evaluate. Always the full name, AgenticGrid Pro.