Mirox-Agent

The Mirox-Agent is the on-plant intelligence that collects data from your renewable energy installation, validates it, runs analysis locally, and streams the results to the Mirox-Cloud. It bridges your physical equipment and the platform so you see live production, health, and losses without touching the loggers yourself.

Agent Overview

The Mirox-Agent operates autonomously, handling all monitoring tasks for an installation. It is more than a passive data collector — it actively:

• Collects Data: Connects directly to your loggers, inverters, meters, and storage to pull readings
• Standardizes Readings: Normalizes every source into one consistent Mirox metric vocabulary
• Validates Data Quality: Range, plausibility, and energy-counter checks before anything is stored
• Runs Edge Analytics: Expected power, performance ratio, curtailment tracking, clear-sky baseline, and a day-ahead forecast — computed on-site
• Inspects the Local Network: Discovers and watches the plant's network devices
• Builds the Digital Twin: Feeds a per-plant model that detects component faults and energy losses

The agent can be deployed either on-site at the installation or in the Mirox-Cloud with VPN access to the equipment. For deployment strategy details, see Deployment Options.

Core Responsibilities

Real-Time Data Collection

The agent actively retrieves data from all monitored equipment, source by source.

Vendor-Specific Adapters:

• Named-Device Adapters: Each supported logger, inverter, meter, or storage family has a dedicated adapter built against that device's own interface — Bluelog, SMA Sunny Central, SMA Power Manager, Sungrow, Huawei SmartLogger, Janitza, Phoenix Contact, FREQCON battery storage, and others
• Real Transports: Adapters speak the device's native interface — vendor HTTP/REST web interfaces, WebSocket feeds, a Microsoft SQL historian, S3-compatible file stores, and weather services
• Built On Request: When you bring a device family we don't yet support, we can build an adapter for it — adding a source is a configuration change, not a redesign

Active Collection Approach:

• Proactively pulls data from devices rather than waiting passively
• Collection intervals tuned per device type
• Parallel retrieval across sources, with one source's failure isolated from the others
• Automatic error recovery and retry on transient problems

For the full list of supported devices, adapters, and the data collection pipeline, see Data Scraper.

Edge Analytics

Beyond raw collection, the agent computes higher-value series directly on-site so you can chart them like any other metric:

• Expected Power: Modelled production for each source, used as the baseline for loss and curtailment
• Performance Ratio: How the plant performs against its expected yield, refreshed as new data arrives
• Live Curtailment Tracking: Attributes foregone production to either the marketer or the grid operator when active power is capped
• Clear-Sky Baseline: A theoretical clear-sky production curve kept as a long-term reference
• Day-Ahead Forecast: A next-day production forecast derived from weather data
• Historical Backfill: When a plant comes online or a gap is found, the agent can replay and rebuild past series rather than leaving holes

These analytics are weather-physics based, not machine-learning estimates. For how they are computed and which metrics they emit, see Data Scraper and Metric Collection.

Digital Twin and Health Monitoring

The agent feeds a per-plant Digital Twin — a live model of the installation's component hierarchy (feed-in meters, inverters, combiner boxes, strings, and panels) that the platform uses to evaluate component health and energy losses.

Two Distinct Engines:

• Configuration Analysis: Discovers and validates each string's characteristics — orientation, panel count, inverter clipping, shadowing, and performance. This analysis is triggered manually or at startup.
• Health Watchdog: Monitors component health and computes energy losses automatically every night, per plant. It simulates expected production, classifies each component's state, and separates genuine outages from mere data-collection gaps. It self-heals by backfilling missing days.

What the Twin Determines:

• Whether each component is producing normally, degraded, overproducing, or not producing
• Whether a missing reading is a real outage (counted as a loss) or only a communication gap (not a loss)
• Energy losses bucketed by confidence (high / medium / low), with weather periods such as snow, fog, and dew excluded so they are never miscounted as component faults

The Digital Twin currently analyzes solar PV installations. Wind and battery analysis are Planned.

For the Digital Twin architecture, see Digital Twin. For the user-facing features it powers, see Digital Twin Features, Component Evaluation, and Loss Detection.

Local Network Inspection

When deployed on-site, the agent maps and watches the plant's local network so you know exactly which devices are reachable.

Discovery:

• Sweeps the configured network ranges to find live devices
• Resolves each device's hardware address and looks up its manufacturer
• Probes devices to classify their type and operating system, with AI-assisted identification for devices that simple rules cannot place

Ongoing Checks:

• Reachability, response-time, and service checks (ping, TCP, HTTP) per device
• Device-level health where supported (system, interface, and resource counters)
• On-demand re-discovery and per-device rechecks you can trigger from the platform

For the user-facing view of this capability, see Local Network Inspector.

Local Processing and Resilience

The agent does meaningful work before anything leaves the plant, and keeps running through connectivity loss.

Data Validation:

• Range Checking: Values stay within physically possible and expected limits
• Energy-Counter Checks: Energy counters must increase monotonically; impossible jumps are rejected
• Freeze Detection: A logger returning frozen, repeated values is detected and flagged
• Cross-Source Plausibility: Readings are checked against weather conditions and the plant's known capabilities

Preprocessing:

• Unit Normalization: All readings converted to standard units
• Derived Metrics: Energy integrated from power, plus performance and efficiency series
• Aggregation: Sums and rollups across the component hierarchy
• Timestamp Alignment: Data from different sources lined up in time

Buffering and Recovery:

• Local Retention: Data is held locally during connectivity loss
• Automatic Resumption: Buffered data is delivered once the connection returns
• Gap Detection: Missing periods are detected and rebuilt by backfill

Access Auditing

When the agent operates the on-plant access proxy, it records who reached which device through it. Each access is grouped into a closed session with an AI-generated summary of the activity, then synced to the platform's audit log — sensitive query data is never retained. For the user-facing view, see the Audit Log.

Configuration and Management

Autonomous Operation

Once configured, the Mirox-Agent runs autonomously without continuous external management. It operates from its configuration, handling collection, validation, analysis, and transmission on its own, and it self-heals after outages by backfilling missing data.

Configuration Approach

The agent is configured per plant, defining:

• Sources: The devices to monitor and how to reach them
• Schedules: What to collect, when, and how often
• Analytics: Which edge analytics and network inspection are enabled
• Limits and Thresholds: Operational boundaries for validation and alerting

Configuration is generated and delivered by the Mirox-Cloud, so onboarding a plant is a self-service activity rather than a manual file edit on the device.

Operational Characteristics

Reliability

• Automatic recovery from transient errors
• Independent source collection — one device failing does not affect the others
• Self-recovery mechanisms and automatic restart on crashes
• Graceful behavior under resource constraints

Performance

• Parallel communication across sources
• Per-device tuning, with backoff for unreliable devices
• Batched, compressed transmission to use bandwidth efficiently

Security

• Encrypted, authenticated connection to the Mirox-Cloud
• Secure handling of device credentials on-site
• Permission-controlled configuration and command access
• Minimal exposure of the plant's network to the outside

Integration with Mirox-Cloud

The agent works closely with the platform:

Data Transmission:

• Real-time metric streaming to the time-series database
• Latest-value updates to the Digital Twin for fast last-value access
• Batched backfill of historical data

Command and Control:

• Live, two-way command channel: stream logs, control individual sources, trigger discovery, and run onboarding probes from the platform in real time
• Remote configuration updates and software distribution

Event Reporting:

• Production-state changes (such as shutdowns and overproduction) reported to the platform as they happen
• Discovered components synced so a plant's component list populates automatically

Related Features

• Data Scraper — the adapters, transports, and edge analytics behind data collection
• Digital Twin — the analysis and watchdog engines that turn metrics into component health and losses
• Deployment Options — on-site versus cloud deployment and what each enables
• Local Network Inspector — the on-plant network discovery and device monitoring view
• mrxnode — the recommended on-site hardware platform for the agent