Modern asset management, enriched with AI. A single source of truth for your assets, their relationships and their history, augmented with accurate metadata.
No. 3No. 2No. 1Reactor coolant loopFTTo volume control tank12345
Figure 1 — Section through the seal package. Injection enters below the No. 1 seal; most of it goes down into the coolant loop and a measured few gallons a minute pass up across the faces to the volume control tank. That upward flow is the only measurement anybody has of the seal.
A healthy modem holds all four steady. Blinking internet with steady power is the reboot case.
Hold the recessed reset for three seconds, then leave it alone: the lights cycle twice before the tunnel comes back. If the readings have not resumed in five minutes, the SIM has lost its provisioning — that is HOW-BAT-006, not a second reboot.
02Close the loop
From alarm to fix in seconds.
Enrich your tag namespace, bring in telemetry and manuals from every vendor, then diagnose.
01
Name· Atlas
Atlas organizes tags, assets, and topologies into a single source of truth that everything else reads from.
Fleet inspections, trouble codes, bay throughput, parts and warranty
DTC-P0420
03Integrations
Your data already exists. Millwright aggregates it.
The historian, the EAM, the procedures library, the shift channel. Millwright connects them into one picture, writes back where you allow it, and leaves every closure to a person.
A marketplace of connectors, and an SDK for the systems only your plant has.
Connected sources
Where this organization's readings, jobs and documents come from.
Cardinal PIAVEVA PI System·Backfilling 90 days of history for the 2R19 review12,480 tagsSyncing
Cardinal EAMWrite-backIBM Maximo·Work orders both ways, closures held for a personConnected
Station proceduresSharePoint·3 of 11 libraries returned 403Degraded
Scoped access, a complete audit trail, encryption throughout, and your choice of where it runs. Check any of them in a demo.
Role-based access
Operator, engineer, planner, contractor, stakeholder. Roles come from your identity provider over SSO.
Fine-grained scoping
Permissions follow the asset hierarchy, down to a single unit, system or point.
Read-only by default
Every connector starts read-only. Write-back is per system, per field, and off until you turn it on.
Audit logging
Every action, access and approval is recorded, queryable and exportable.
Encryption
TLS on every hop, including to the historian. At rest with keys you can bring and rotate.
Cloud or on-premises
Hosted, in a single-tenant VPC in your own account, or on the plant floor next to the historian.
Never trained on your data. Nothing you is used for training.
Always approved by a human. Approving and closing stay human-controlled.
05One picture, one purchase
Everyone on the same page.
Four tools means four rollouts, four sets of credentials and four versions of what happened last Tuesday. The fault you are chasing usually lives in the gap between two of them.
Field engineers
Work the fault on a shared board with the agent. It brings the evidence, you bring the judgment.
Plant operators
Start the shift on the overview: output, capability, active alarms and open work.
Customers and stakeholders
Follow a site without standing in the control room, scoped to what they are allowed to see.
Our historian has twelve thousand points and three naming conventions. What actually happens?
A resolution pass reads point names, descriptors, units, ranges and whatever hierarchy exists, and proposes an asset for each point with a confidence you can see. Expect roughly one point in six to come back ambiguous, undescribed or long dead. Those go to an engineer rather than being guessed at, and each correction applies to the rest of the site. This is the part with real hours in it, and we would rather you budget for it than be surprised.
Does our data leave the plant?
We offer both cloud and on-premises options. Data is only transferred if you decide it should be. Run it hosted, in your own cloud account, or on-premises next to the historian. Model inference is the one step that can cross the boundary, and where it runs is a setting rather than an assumption. Nothing you connect trains anything.
What happens when the agent is wrong?
Every claim carries its evidence: the telemetry, the ticket, and the support article behind it, so a conclusion you disagree with is one you can see the working for and delete. It cannot close a ticket, write back a closure or sign anything. And where the telemetry does not decide, the run stops and asks rather than picking.
Does this add to our alarm load?
It should reduce it. Related alarms group into one investigation, so a flood arrives as one row rather than nine, and Millwright raises no alarms of its own. It does not replace your alarm management system or write to the DCS.
Who has to sign off to run a pilot?
For the first one: whoever can approve read-only access to the historian, plus an engineer with a few afternoons for tag review. We keep the opening scope inside one person's authority on purpose. A rollout that needs the historian, the CMMS, the procedures library and the control room to all agree first is not a pilot, it is a program, and programs are how the last ten years went.
How long does a first deployment take, and what do we need to have ready?
Weeks rather than quarters for one system, read-only. The long pole is tag review. What it needs from you: read access to the historian, somebody who knows the plant well enough to correct a mapping, and somewhere to run. What it does not need: a data migration, a change to your CMMS, or anything touching the control system.
See it running on your own plant.
We are onboarding a small number of operations teams. Three questions, and we will be in touch.