AI agents for any EAM system that run the maintenance loop.
Governed AI maintenance agents for IBM Maximo, Nexaform EAM and any other EAM. They watch your assets, plan the work, prepare the parts and draft every action. Your team approves. The agents learn.
- When
- Tonight 00:45–04:00 · SEC06 possession
- Crew
- S&T Night Crew B
- Parts
- Detection contacts ×2, lubricant kit ×1 · reserved
Predictive maintenance alerts were only step one
Predictive analytics stops at the alert. Agentic asset operations go further: agents plan the job, parts and schedule, act in your EAM within approved limits and learn from every outcome.
- 1Detects a pattern in the data
- 2Raises an alert
- 3Waits for a person to notice
- 4A planner builds the work order by hand
- 5Learning stops at the alert
- 1Senses conditions continuously
- 2Reasons over history, manuals and context
- 3Plans the work: job plan, parts, schedule
- 4Acts in your EAM within approved limits
- 5Learns from every outcome
Alert overload
A manual last mile
Poor data in
No learning loop
How AI maintenance agents work: sense, understand, decide, act, learn
Predictive analytics, anomaly detection and failure forecasting are built in. They are tools the agents use, not another dashboard to watch.
- 01
Sense
Reads meters, sensors, alarms and work history around the clock.
- 02
Understand
Puts signals in context: asset history, manuals, failure modes.
- 03
Decide
Weighs risk, cost and schedule, and picks the best next action.
- 04
Act
Drafts work orders, reserves parts and books crews in your EAM.
- 05
Learn
Feeds every outcome back so the next decision is better.
Seven AI maintenance agents, one coordinated team
Condition agent
Reliability agent
Planner agent
Parts agent
Technician assistant
Data steward agent
Orchestrator
Coordinates the agents, applies your policies, routes approvals and records every decision in the audit trail.
Create your own Perimo agents
In the upcoming version, users can build their own agents and run them from Perimo Copilot with a command such as /dead-stock.
One issue, from first signal to closed work order
- 02:10Condition agentBearing temperature on pump P-104 drifts outside its learned normal range.
- 02:12Reliability agentMatches the pattern to past seal and bearing failures and rates the risk high.
- 06:30Planner agentDrafts a work order with job plan and crew, slotted into Thursday’s planned stop.
- 06:31Parts agentConfirms the bearing kit is in stock and reserves it against the work order.
- 07:05Maintenance plannerReviews the evidence and approves in the EAM with one click.
- ThuTechnician assistantGuides the repair and records problem, cause and remedy correctly.
- AfterAll agentsThe outcome updates the models, and similar pumps are re-scored.
Illustrative example, synthetic data.
AI agent autonomy levels: you decide how much the agents do
Agents analyse and explain. No actions are proposed.
Agents suggest actions. People create the work themselves.
Agents prepare complete work orders and requests. A person approves each one.
Agents complete low-risk, reversible actions within policy, such as reserving parts.
Safety-critical assets always need human approval. Autonomy is set per action type, asset class and site, and can be withdrawn at any time.
AI agents that work inside the EAM you already run
They act through each system’s own APIs, workflows and permissions, and never see or do more than the approving user is allowed to.
Swipe sideways to see more →
Two AI maintenance use cases, from EAM data to outcome
Both come from metro rail, where a single failure in the morning peak can hold dozens of trains.
Stop the point failure before the morning peak
Point machine PM-23’s throw time has drifted from 3.80 s to 4.28 s. The EAM alarm is set at 5.0 s, but similar machines failed between 4.6 and 4.9 s. The agents catch the trend, draft the work order for tonight’s possession and reserve the parts. A planner approves at 07:05.
- When
- Tonight 00:45–04:00 · SEC06 possession
- Crew
- S&T Night Crew B
- Parts
- Detection contacts ×2, lubricant kit ×1 · reserved
- Scope
- All 8 doors: guide rails, rollers, controller, functional test
- When
- Thursday, with the planned exam · RS Depot Crew 1
- Parts
- 8 roller sets reserved · guide-rail kit on draft PR
Train doors that keep coming back
Car TS-07-C2 logs four times the fleet’s door faults, and 40% of door work orders have no failure code. The agents read the messy history, find the car-level pattern and raise one investigation covering all eight doors. No new sensors were needed.
Illustrative example, synthetic data.
Every job makes the AI agents more useful
- 01
Capture
The technician assistant records correct failure codes and notes at close-out.
- 02
Clean
The data steward fixes past codes, duplicates and hierarchy gaps.
- 03
Learn
Risk and condition models retrain on better data every cycle.
- 04
Trust
Higher acceptance lets you raise autonomy where it is earned.
AI agent governance built for the controls your auditors expect
Your policies decide
Scope, limits, approvers, time windows and a kill switch, versioned like any controlled change.
Every action proposed first
Approvals are bound to the exact action, single-use and time-limited.
Tamper-evident audit
An append-only, hash-chained record of every decision, replayable and exportable to your SIEM.
Security from identity to data
SSO with MFA, least-privilege agent identities, encryption and private connectivity.
Grounded AI
Every number comes from your systems. When data is missing, the agent says “cannot determine”.
Deploy anywhere
Our cloud, a dedicated cloud, your cloud account or on-premises, with your choice of AI model.
Start in shadow mode: agents advise, people decide
Proof on your own data before any agent drafts real work.
- Weeks 1–2
Connect
Read-only access to your EAM and condition data, plus a data quality baseline.
- Weeks 3–8
Shadow
Agents propose actions on live work while your team works as normal.
- Weeks 9–10
Review
Acceptance, accuracy, early catches and time saved, measured in your own numbers.
- Week 11 onward
Go live
Switch on approval-based drafting for agreed sites and action types.
How Asset Agents compare: built for multi-EAM estates and governed autonomy
| EAM-suite agents | Industrial AI platforms | Perimo Asset Agents | |
|---|---|---|---|
| Works with | Their own EAM | Through integration projects | Any EAM, inside its own workflow |
| Starting point | Latest suite version | Data-platform rollout | Your existing EAM data |
| Autonomy control | Vendor-defined | Varies | Your policy per action, asset and site |
| Audit trail | Within the suite | Varies | Tamper-evident, exportable to your SIEM |
| Best fit | Single-vendor estates | Data-heavy process plants | Multi-EAM, mid-size to large estates |
Swipe sideways to see more →
Generalised comparison of product categories; individual products differ.
Questions about AI agents for Maximo and other EAMs
Do Perimo Asset Agents include predictive analytics and anomaly detection?
Yes. Predictive analytics, anomaly detection and failure forecasting are built in. The agents use them to turn signals into planned, resourced work orders instead of another alert.
What is an AI agent for IBM Maximo?
An AI agent for Maximo is software that reads Maximo data such as meters, work history and failure codes, reasons about what should happen next, and prepares the action in Maximo itself: a drafted work order with job plan, labour and parts. With Perimo Asset Agents, a person approves each action by default.
Do the agents change our EAM without approval?
Not by default. At Level 2 every work order and request is drafted for a person to approve. Level 3, where agents complete low-risk, reversible actions on their own, is opt-in per action type and never applies to safety-critical assets.
Which EAM systems do Asset Agents work with?
IBM Maximo, including 7.6 and MAS 9.2, and Nexaform EAM. Other EAMs such as SAP connect through connectors built on each system’s own APIs and permissions.
Do we need new sensors?
No. Agents start with the work history, meters and failure codes already in your EAM. Condition data from historians adds more signal when you have it.
Which AI models do you use, and is our data used for training?
Model providers are configurable per customer: approved hosted models, models in your own cloud account, or on-premises models. Your data never trains shared models without your explicit written consent.
See the AI agents work on your own maintenance history
Bring a few months of work orders from Maximo or another EAM. We will show you what the agents would have proposed.