Machine Agents
AI agents that monitor industrial sources (sensors, PLCs, SCADA) and automatically respond to events with analysis, tasks, and notifications.
Machine Agents
Machine Agents are the core of industrial monitoring in Rela AI. When a sensor detects abnormal vibration, when a PLC generates an alarm, or when the SCADA reports a critical event, it is a Machine Agent that receives that event, analyzes it with artificial intelligence, and decides what to do: create a task, notify the technician, escalate to the supervisor, or a combination of all three.
What is it for?
In most plants, machine events generate thousands of alarms per month. The problem isn't lack of data — it's the lack of timely response. A critical alarm that nobody sees at 3 AM can become a costly failure by dawn.
Machine Agents solve this with intelligent automation:
- Analyze each event in the context of the equipment's history
- Create tasks automatically and assign them to the correct department
- Notify the technician via WhatsApp or email at the moment of the event
- Escalate the alert if nobody responds within the defined time
How does it work?
A Machine Agent connects to one or more event sources (sensors, PLCs, SCADA) through HTTP (webhook), MQTT, or OPC UA protocols.
When an event arrives, the agent:
- Verifies it exceeds the configured minimum severity
- Applies throttle — prevents the same event from being processed repeatedly
- Sends the event to the AI model with the context defined in the prompt
- The model analyzes the event and decides the response
- Executes configured actions: create task, send notification, activate escalation
All agent responses and actions are recorded in the event history.
How to use it?
Create an agent
- Go to Alarms in the sidebar.
- Click Create Agent.
- Assign a descriptive name — for example: "Agent-Compressors-LineA" or "Boiler-Monitor."
Configure the agent prompt
The prompt defines how the agent thinks and acts. It is organized in 4 sections:
| Section | What to write |
|---|---|
| Identity | Who it is and what equipment it monitors. Example: "You are the monitoring agent for Line A compressors. You have knowledge of equipment C-01, C-02, and C-03." |
| Instructions | How to analyze events. Example: "When you receive a vibration alarm, evaluate severity by comparing with history. If vibration exceeds 8 mm/s, classify as critical." |
| Rules | Limits and criteria. Example: "Don't create tasks for INFO severity events. Only escalate if the event is still active after 30 minutes." |
| Tools | Auto-generated when tools are assigned. |
Use the Enhance with AI button to have the system optimize the prompt for each section.
Connect event sources
In the Event sources section, click Add source and configure:
- HTTP (Webhook): generates a unique URL where your SCADA or system sends events via POST
- MQTT: connects to an MQTT broker with the equipment topic
- OPC UA: connects directly to a PLC or industrial OPC UA server
See detailed documentation at Event Sources.
Configure automatic task assignment
When the agent detects a relevant event, it can create a task automatically:
| Field | Description |
|---|---|
| Department | Which team gets the task (required) |
| Assigned to | Specific person or "automatic" (system selects nearest personnel) |
| Priority | Initial urgency of the task |
Configure automatic notifications
The agent can notify immediately upon detecting an event:
- By WhatsApp: select a connected number and recipient (person or custom number)
- By Email: select an active email account and recipient
Configure escalation
Define what happens if nobody addresses the event within a certain time:
- Define the minimum severity to escalate (only Critical and Emergency, for example)
- Add escalation steps with:
- Step name (e.g., "Notify supervisor")
- Wait time before executing (e.g., 30 minutes)
- Notification channel (WhatsApp, Email)
You can add multiple steps: technician at 15 min → supervisor at 30 min → plant manager at 60 min.
Assign tools
Select the tools the agent can use when processing events: database queries, task creation actions, external API connections, etc. See Tools.
Key benefits
- Automatic response to critical events without initial human intervention
- Intelligent analysis with context — not just simple rules, but reasoning
- Guaranteed escalation — no critical alarm ever goes unattended
- Complete record of every response for subsequent analysis and compliance
- Integration with multiple industrial protocols without complex network configuration
- Automatic assignment to the nearest technician or correct department
Common use cases
Scenario 1: 24/7 compressor monitoring The "Compressor-Monitor" agent is connected to 4 OPC UA sources (one per compressor). At 2:30 AM, Compressor C-02 generates a high temperature alarm (92°C, limit 90°C). The agent processes the event, evaluates the context (temperature has been rising for 15 minutes), classifies it as "critical," creates a task in the Mechanical Maintenance department, and sends a WhatsApp to the night technician: "CRITICAL ALARM — Compressor C-02: temperature 92°C (limit: 90°C). Task #MAN-0234 created. Confirm you are attending." If the technician doesn't respond in 20 minutes, the next escalation step notifies the plant supervisor.
Scenario 2: Intelligent alarm filtering During a shift, the SCADA sends 45 events. The agent automatically filters: 38 are informational (INFO level) and records them without action. 5 are warnings (WARNING) and records them with analysis. 2 are critical and creates tasks for each. The technician only receives 2 notifications during the entire shift — the ones that matter.
Scenario 3: Specialized agent by area The plant has 3 agents configured: one for the production line, one for power systems (generators and UPS), and one for water treatment systems. Each has a specific prompt with knowledge of its equipment, and events go to the correct agent based on the source.