At plants with battery storage, curtailed energy and the compensation shown against it now come from what the inverters produced, and fall where they read too high. The feed-in meter sits behind the battery, so quarter-hours spent charging looked like no production at all and were claimed in full.
An organization's VPN row now has a "Move to another host" action. Before the move is confirmed, it shows where the VPN would land, how many plants restart on site and lose data briefly, which plants get a new VPN address, and which routers keep the old one and stay disconnected until restarted.
Curtailed energy and the compensation shown against it now fall at plants whose grid connection is smaller than their inverters. A limit from the direct marketer that merely restates full output was counted as curtailment all day long and is now judged against the smallest of panel peak, inverter rating and grid connection.
Inverters now show what they report about themselves: "Stopped by command" when one was switched off, "Protective stop" when its protection tripped, and "Device fault" when the machine reports a fault the measurements cannot show. The maker's own wording and how often it was raised are shown alongside.
The status code an inverter reports about itself now opens an event when the device was stopped by command, tripped on its protection, raised a safety alarm or is throttling itself back. The health column shows "Stopped by command" and "Protective stop", and every code one inverter reports is held in a single event.
The revenue tables now show an "Energy (§ 51 applied)" column: the energy each band's Marktprämie is actually paid on. Each band's premium is now worked out from that energy instead of being shared across the bands by capacity.
A swapped inverter, listed twice because the data logger gave the new unit a new address, is now spotted as a pair and offered for confirmation on the components page. Once confirmed, the new inverter takes over the old one's configuration, the old unit is retired, and the graphs keep the earlier history.
Each remote-access connection on a plant's VPN tab now shows whether it is really working, rechecked while the tab is open. One that is set up but not running is marked red, with the next step to take, and one that is running while the remote site has not connected is marked amber.
Mischpreis can no longer be chosen under a contract's "Revenue model". A contract still set to it is valued at "Marktwert (as-if-produced)", and compensation for curtailment ordered by the marketer follows that model from this release, while figures already issued keep their value.
Action required: Plant and portfolio managers: open the market configuration of every plant whose direct-marketing contract was settled on the Mischpreis model, and confirm that "Marktwert (as-if-produced)" or "Fixed value" is the right revenue model for that contract. Marketer-curtailment compensation is calculated on the chosen model from this release; support helps you with an affected contract.
The comparison of settlement methods on the revenue page now uses the figures from the monthly settlement. A month that has not been settled yet reads as unavailable instead of showing a number.
For plants settled on the flat-rate method, a curtailment is now measured against the power in the last valid period before it, rather than the highest power found earlier. Curtailed energy and compensation fall wherever the old rule lifted them, while figures already issued keep their value until a month is rebuilt.
Where the data a plant's settlement method needs has not been published yet, lost energy and compensation worked out a different way were shown as the plant's own. The row and its dialog now read "Satellite data not yet available" instead, or "No usable measurement data" where the gap is permanent.
A quarter-hour excluded from remuneration by § 51 EEG no longer shows a euro amount. The amount disappears when the month is recalculated, while the excluded energy stays in kWh.
An inverter at full output but showing wear is now flagged with "Elevated temperature", "Insulation weakening", "Phase imbalance" or "Efficiency declining". Each problem is marked on the component and listed separately.
Remote access over IPsec (IKEv2), for site routers that speak nothing else, now shows whether the remote site is connected, with its availability and traffic, instead of an unknown state. A site set up but never connected stays in the list, and "Add peer" attaches a second router.
The assistant now answers where a setting or page is with a link to it, and explains what a type of event means or how a measurement is worked out. It also lists what a plant measures, draws the curve for a component or a whole plant, and points out stretches with no data.
Curtailment running into the night no longer counts energy from the irradiation sensor's night-time reading, a quarter-hour below 5 W/m² now counting as zero. Curtailed energy and the compensation shown against it fall at plants settled from measured irradiation, for instructions reaching into the evening or early morning.
A "Curtailments" section on a plant's "Production" page now lists every settled curtailment and every live detection in the period shown, each with its window, its Ausfallarbeit, how it was settled and the compensation. The graphs now draw the same figures, so the last days of a period stay empty until the settlement has run.
The "Add a VPN connection" wizard now offers "IPsec (IKEv2)" beside WireGuard and OpenVPN, for industrial routers that speak nothing else. Mirox can host the tunnel for the plant or connect to a router that hosts it, with a pre-shared key that is generated or pasted.
Compensation for curtailed energy is now shown only for quarter-hours from 23 December 2025 on, the day that rule took effect. A compensation figure for a delivery month lying wholly before that date disappears the next time that month is settled.
Every network of a plant can now be tagged by what it carries — "Plant control", "Network infrastructure", "Cameras" or "Security systems" — when it is added, edited or opened for remote access. The new "Security Officer" role reaches only the networking, camera and security networks, never production, revenue or plant control.
A new "Control Units" tab on a plant's components page now lists the signals the plant is regulated by: the cap from the grid operator, the cap from the direct marketer, and the setpoint the plant controller applies. Each signal names the data logger reporting it and can be renamed, hidden or given a note.
The "VPN auto-heal" wizard can no longer be started, while a run started earlier still opens. A plant's VPN connection is now diagnosed from "Inspect VPN connection" on the plant, where "Start AI diagnosis" takes the router on site and what is needed and lets the assistant read the configuration and the logs itself.
Action required: Anyone who used the "VPN auto-heal" wizard: start the diagnosis from the plant instead. Open the plant's networking page, choose "Inspect VPN connection" on the VPN, go to its diagnosis step, name the router on site and what you need, and carry on with the assistant in the tab it opens. Starting a new auto-heal run is refused from this release; runs you started before it stay readable.
Curtailed energy and the compensation shown against it rise for periods before mid-May 2026: quarter-hours that carried no measurement now count, worked out from the plant's own irradiation sensors and feed-in meters. Plants settled on the flat-rate method get figures there for the first time.
The components page now has a tab for each battery level — "PCS", "Box", "Storage", "Module" and "Cell" — with a checkbox on every row. "Edit Components" then sets or clears the datasheet for the ticked rows, for everything below the filtered parent, or for every component of that type.
A curtailment episode excluded with "Exclude from settlement" is now left out of the month's curtailed energy and compensation. Until this release the exclusion was recorded, but the month was settled as though nothing had been excluded.
An organization's own Anthropic, OpenAI or self-hosted account can now be added by an administrator or moderator under "AI providers" in the organization's AI settings. The models members may use are picked from it and one marked as the default, after which members choose a model per conversation.
Battery plants now have an "Energy flow" page. It shows live power through the converters and racks, state of charge, the voltage and temperature spread of the cells, a heatmap of every cell in a rack, and the same curves over a chosen period.
A limit from the grid operator or the direct marketer left between 85 % and full power no longer counts as a curtailment, even where it does curtail. Curtailed energy and the compensation shown against it drop at plants whose limit sat in that band, at some for a whole month.
With "Group string names by MPPT" switched on under a solar plant's "Component Naming" settings, its strings read "MPPT 2 – String 1" in the components list, a string's details, the analysis tables and tickets. Strings whose inverter does not report how many inputs each MPPT has keep their plain numbering.
The assistant now answers what a plant earned in a month or a year: production revenue, the compensation for grid and marketer curtailment, and the market value behind them. Several plants are compared side by side and against the yield the plan expected, and months not yet settled are named as provisional.
A model can no longer be chosen for the assistant, which always answers on the model Mirox provides. A model an organization had chosen earlier is cleared the next time the AI settings open.
A dropdown beside the chat's input bar sets how much reasoning the assistant does before it answers: "Deep", "Medium", "Low" or "Off". Every conversation opened or switched to starts on "Medium".
A feed-in meter's own reading is used again on bright days, and the day counts as "measured" instead of "integrated". Before, a high reading was thrown out as impossible and replaced by the total added up from the inverters, or by zero where there was no substitute.
Each inverter in the monitor's connection list now leads with the reason it cannot be reached: an open outage on the network device its data logger sits behind, naming that logger. Such an inverter counts as a connection issue at once, instead of only after an hour without a signal.
A plant's open-event count no longer adds the grouping alert on top of the dead components themselves: twelve dead inverters read as twelve.
Every event in a plant's activity list, dialogs and tickets now shows its name in the user's language instead of an untranslated one.
The month chart no longer marks a day as a connection outage when the plant produced right through a brief drop, or before monitoring began.
A Solar-Log logger can now be stopped, under its settings, from reporting a feed-in meter another source already reads, which counted grid power twice.
In the API, the site list can be narrowed by portfolio and type, both comma-separated, with the total and the paging following the filter.
The monitor list now has a "Live" column showing the power each plant is feeding into the grid, refreshed every 10 seconds. Below it stand the plant's two power limits, from the grid operator and from the direct marketer, each coloured only while it is holding the plant back.
Curtailment episodes detected at a plant now appear in the plant's events as their own entries, with the limit that applied and the period covered. "Exclude from settlement" on an entry leaves it out of that month's curtailment figures, and "Re-include in settlement" puts it back.
"Reachable subnets" now shows where two plants' networks overlap, not only where they are identical, and lets the plant serving the shared addresses be chosen.
On a per-Wp day graph, the reference curve now follows the selected components, so an east-west plant shows both of its peaks.
The sign-in page now states how long the pause after repeated failed sign-ins lasts, and that changing the password does not end it early.
Where two sources sent readings for the same minutes — after a restart, for example — a plant's curtailed energy for that quarter-hour was counted twice, and the compensation shown read too high. Such a quarter-hour is now left out of the settlement, and figures already issued keep their value.
The "Edit logger" dialog now has a "Configuration" tab. There, the options a data logger offers and the "Serial / metric id prefix" its components are identified by can be changed without deleting and re-adding the logger, and "Apply changes" sends them to the plant.
A plant's expected monthly yield now allows for panel ageing counted from the day the plant was built, not from the day Mirox started collecting its data. Expected figures fall where the plant is older than its data, and no ageing is applied where no build date is recorded.
Compensation for curtailed energy is now priced against the official published value. Until that value is published, about two months after the delivery month, the figure stays an estimate and is corrected later.
An inverter, a string or a combiner box opened from the analysis page showed its produced, expected and lost energy a thousand times too high — kWh labelled as MWh — and the battery overview showed DC power the same way. Both read correctly again.
mirox 2.15.166 – 2.15.172
Web appImprovements and fixesNetwork & remote access
A solar plant without a feed-in meter can now be given a virtual one with "Add Virtual Meter" on the plant's "Feed-in Meters" tab. The on-site agent then reports the plant's grid power and total grid energy as the sum of its inverters, as estimates rather than measured readings.
A field cleared in an organization or contact dialog now saves as empty, and an unchanged dialog saves nothing.
Through the API, /v1/sites now lists the reachable plants without health figures, and /v1/sites/status returns health for the plants named in the call.
A remote-access route whose subnet contains the reserved address 10.112.0.1 is now flagged, and only that single address stays unreachable inside the plant.
A "Status" column on the monitor page shows a green checkmark when all is well, otherwise how many problems are open and how serious they are. Selecting it opens "Active events": everything still going on at that plant, grouped by type, each row linking to the part it concerns.
The "Sites" tab now lists, under "Shared with your organization", the plants and portfolios other organizations share, below the organization's own. Each row names who shared it, the role given, the portfolio a plant comes through, and reads "visible only — no data access" when it cannot be opened.
A raw data export is refused if one column would hold more than 36,000 values — a full year at 15-minute resolution still fits — or if splitting by plant would give more than 50 columns, one per metric and plant. A shorter period, a coarser step, or merged plants keep an export within both limits.
Action required: API users: keep every raw data export within 36,000 values per column — a full year at 15-minute resolution — and, where you split by plant, within 50 columns (metrics × plants). Anything above either limit now fails instead of running. Shorten the time range, choose a coarser step, or merge the plants instead of splitting them.
The assistant now answers a question about a plant's remote access with every connection: its role, protocol, the sites it connects, the networks it routes, and each one's last known status. Where something looks wrong, it names a likely cause.
Three tabs beside the logs — "Live status", "Diagnose" and "Debug" — show whether a plant's remote access is working. "Live status" shows each connection's address, throughput and how long ago it last checked in, updated once a second.