Product · how it works

One incident, from alarm to verified fix.

Astyr is not one model answering once. It is a team of specialized agents with the operator in the loop, and everything is grounded in the plant's own documents. Here is the exact loop — the same one you can run live below.

01Multi-agent orchestration

The Watchdog detects. The Orchestrator routes. Specialists resolve.

A Watchdog watches every live feed; on a fault it hands the incident to the Orchestrator, which routes typed state to specialist agents in order — correlation, root-cause, responder-matching, validation, remediation, and cost/inventory/dispatch.

Correlationfuses the alarm storm with the network topology to isolate the one site that matters — and its equipment class.

02Documentary grounding

Every claim is tethered to your own documents.

Astyr ingests the plant's real, messy corpus into a clean spatial and technical knowledge base. Below a confidence threshold, the root-cause agent refuses to guess and re-retrieves matching-signature past breakdowns; if it is missing a document it needs, it asks the operator for that specific one.

Plant layouts & topologyElectrical schematics & P&IDsEquipment manualsPast breakdownsSafety & lockout proceduresSpare-parts catalog

Agent claims

Rectifier module PN-RECT-48-2000 is the fault origin

Site draws from the −48 V DC power plant on feed B

Swap procedure requires an isolation step first

Carrier's own docs

Rectifier Maintenance Manual

cited · p. 42

PAR-021-NORD site dossier

cited · p. 7

Field Repair Procedure R-118

cited · § 3.2

03The validation loop

Verified by the operator before anyone touches the machine.

On confirmation, the incident flows to a cited remediation procedure and a single cost / inventory / dispatch agent. On refusal, it pivots back to re-diagnose. The dangerous failure mode in a safety-critical setting is confident-wrong — Astyr is architected against it.

confirmed → remediate refused → pivot & re-diagnose human field veto

It never acts on an unverified diagnosis.

The loop forces a real field measurement before any prescription is produced. Astyr hands the operator a hypothesis and a safe test — not an instruction to act.

It knows when it does not know.

Confidence-gating and refuse-to-guess, an explicit request for the specific missing document, and graceful degradation to named manual triage rather than a fabricated answer.

It augments — it decides nothing physical.

The human verifies and acts, with cited safety and lockout procedures surfaced before movement. Astyr's recommendation sits under the operator's sign-off.

04See it live

Watch one incident — then run it yourself.

See Astyr live on one critical-infrastructure domain — telecom-site energy fault response — illustrating the same architecture we bring to manufacturing. Watch the replay, then unlock the control room to run one end to end.

Unlock the interactive control room

ACCESS · GATESTATUS · LOCKED

Live control room

Run a real incident end to end.

Detection, grounded diagnosis, the operator's field validation, and a cited action report — the exact multi-agent loop, live in your browser.

This build runs one critical-infrastructure domain — telecom-site energy fault response — illustrating the same architecture we bring to manufacturing.

Watch the agents fan out from a single alarm
Confirm or refuse a diagnosis from the field device
Download the cited action report

Astyr · demo

Want it grounded on your plant's real documents? That is exactly how a pilot starts.

Request access