Solar monitoring
AI agents for solar monitoring
An AI agent that reads your monitoring portals, triages every alarm and raises only the ones that need a person, with the evidence.
FusionSolar6 plants11:452 of 38 need a person
- 2
Needs a personPlant 1 · INV-04, insulation at 3.8 kΩ. Plant 4 · CB-11.3, two strings at 0 A.
- 7
Already in a work orderWO-2279, WO-2284 and five more, all on schedule.
- 29
Cleared as noiseCommunication blips, dew-time insulation dips and grid frequency events.
01In short
What does an AI agent for solar monitoring do?
Invertix’s Monitoring agent is an AI agent for solar monitoring. It reads your existing monitoring portals, such as FusionSolar, the way an operator does. It triages every alarm and raises only the ones that need a person, each with the cause and the evidence: the fault, the reading and the energy at stake. After a repair, it checks the plant in the portal before the work order closes. It does not replace your monitoring system.
Invertix works with Altenia, Terna and TotalEnergies On. The energy companies we work with manage more than 50 GW in total: their combined portfolio, not capacity operated by Invertix.
Last updated
| What it reads | Your monitoring portals, such as FusionSolar, the way an operator does |
|---|---|
| What it does with each alarm | Triages it, and raises only the alarms that need a person |
| What it posts | The cause and the evidence: the fault, the reading and the energy at stake |
| Where alarms appear | In your Microsoft Teams channels |
| After a repair | Checks the plant in the portal before the work order closes |
| How it signs in | With logins kept in an encrypted secrets vault, never visible to the models or to Invertix staff |
| What needs your approval | Any change to plant controls |
| Record | Every agent action is logged and visible to your team |
| Works with | Altenia, Terna and TotalEnergies On |
| Scale | 50+ GW managed by the energy companies we work with: their combined portfolio, not capacity operated by Invertix |
| Company | Invertix GmbH, Munich, Germany. Backed by Y Combinator, Vireo Ventures and Italian Founders Fund. |
02Alarm triage
How does it decide which alarms need a person?
It reads every alarm the way your operators do, clears the noise and raises the few that need a person, with the evidence.
On the demo fleet, 38 alarms come in across six plants. The Monitoring agent finds the two that need a person, sees that seven already belong to a work order, and clears 29 as noise.
The film shows 38 alarms from FusionSolar on a demo fleet of six plants. The Monitoring agent sorts them: two need a person, seven are already in a work order and 29 are cleared as noise.
In detail about alarm triage
Needs a person. On the demo fleet, a low insulation resistance on INV-04 at Plant 1 and two strings at 0 A at Plant 4. Each goes to your team with the cause and the evidence.
Already in a work order. Seven alarms belong to work orders that are already open and on schedule.
Cleared as noise. 29 alarms need no action: communication blips that are back within five minutes, insulation dips at dew time that recover by mid-morning, and grid frequency events.
Your rules. For each workflow, you choose whether the agent works fully autonomously or a person verifies each action.
03Worked example
What does an alarm look like when it reaches the team?
A post in your team’s Microsoft Teams channel, with the cause and the evidence: the fault, the reading and the energy at stake.
When the alarm needs a technician, the Dispatch agent takes it from there. After the repair, the Monitoring agent checks the plant in the portal and posts the result before the work order closes.
In detail about the example
The cause. In the example, the agent names the likely cause, a damaged cable rather than dew, before it asks anyone.
The evidence. The fault, the reading (3.8 kΩ against a 50 kΩ limit) and the energy at stake (1.8 MWh by sunset), in one card.
The hand-off. The Dispatch agent can propose the nearest free qualified technician and, once your team approves, book the visit.
The check. After the repair, the Monitoring agent checks the plant in FusionSolar: grid-connected at 293.6 kW. Then the work order closes.
04Access
How does it access our portals?
With logins kept in an encrypted secrets vault. The logins are never visible to the models or to Invertix staff.
Your people sign in with single sign-on, and roles and permissions decide who sees and does what. Every agent action is logged and visible to your team, and agents write to plant controls only with your approval.
| Portal logins | Kept in an encrypted secrets vault |
|---|---|
| Who sees the logins | Neither the models nor Invertix staff |
| Your people | Single sign-on, roles and permissions |
| Plant controls | Only with your approval |
| Record | Every agent action is logged and visible to your team |
| AI models | OpenAI, Anthropic and Mistral, always through the Invertix harness |
| Names | Site names and people’s names are replaced by placeholders, such as [SITE_1], before any model sees a request. |
| Hosting | EU cloud by default, encrypted in transit and at rest |
05Your monitoring platform
What does an AI monitoring agent add to a monitoring platform?
Your monitoring platform stays the system of record. The agent does the work that follows an alarm.
Your monitoring platform collects the plant data and raises the alarms. The Monitoring agent works inside it: it triages the alarms, posts the cause with the evidence, hands the repair to the Dispatch agent and checks the fix. Invertix does not replace your monitoring platform, SCADA or work-order system.
- 01Your platform
Collects the plant data and raises the alarms, as it does today.
- 02The Monitoring agent
Triages every alarm and posts the cause with the evidence in Microsoft Teams.
- 03Your team
Decides what needs a person, and approves before plant controls change.
In detail about your monitoring platform
Connecting a system. Our engineers connect each system with your team, as part of the deployment.
One operational model. Invertix brings the records of your portals, Microsoft Teams and work-order system into one operational model: a single, shared record of your plants, alarms, work orders and people.
Beyond monitoring. The same alarm can go on to the Dispatch agent, which can propose a technician and, once your team approves, book the visit, and to the Asset management agent, which logs the fault to the inverter’s fault history. AI agents for solar O&M →
06Questions
Questions about AI agents for solar monitoring.
Does it replace our monitoring portal or SCADA?
No. The Monitoring agent works inside the monitoring portals you already use, such as FusionSolar.
Where do the alarms appear?
In your Microsoft Teams channels, with the evidence attached: the fault, the reading and the energy at stake.
What happens after a repair?
The Monitoring agent checks the plant in the portal before the work order closes.
How are portal logins protected?
They sit in an encrypted secrets vault that neither the models nor Invertix staff can see.
Can the Monitoring agent change plant settings?
Only with your approval, whatever a workflow’s setting.
Can we see what the agent did?
Yes. Every agent action is logged and visible to your team.
Which AI models see the alarm data?
Models from OpenAI, Anthropic and Mistral, always through the Invertix harness. Site names and people’s names are replaced by placeholders before any model sees a request.
Who sets it up?
Invertix’s own engineers, working inside your operations team. They connect your portals and Microsoft Teams as part of the deployment.
What should we check before giving an agent access to our monitoring portals?
How it signs in and who can see the logins, what it may do without approval, whether every action is logged, which models see the data and in what form, and where the data is hosted.
managed by the energy companies we work with.
Bring us one recent alarm.
Our engineers will walk you through how an agent would handle it, step by step, with your team in control.
Europe +34 606 62 83 84 · US +1 (415) 718-7732