Ask operational questions in plain language.
Explore live and historical data by asking. Every answer comes back with supporting charts, the relevant evidence and recommended next checks — grounded in the same shared model of sites, spaces and assets as every dashboard and rule.
One assistant, every team
Ask in plain language
Type a question the way you would ask a colleague. It is resolved against modelled assets, spaces and history — not against a free-text guess.
Evidence, not just a number
Each answer arrives with the chart, the affected assets and spaces, possible contributing factors and the next checks to run — plus one-tap follow-ups and comparisons.
Insights arrive unasked
The assistant keeps reading the portfolio in the background and surfaces what changed — a drift, an overspend, an asset behaving unlike its peers — as a reviewable insight before anyone thinks to ask.
Just ask
From a question to the evidence behind it
The same assistant handles a quick lookup, a scheduled check and a deeper investigation — here it is, exactly as your team sees it.
Answers with charts, evidence and next checks
Every reply comes with the chart, the affected assets and spaces, and the checks worth running next — plus quick buttons to drill down, notify a team or save a report.
- Possible contributing factors, with the evidence behind them
- One-click follow-ups and comparisons
- Export to PDF or save as a report
May is tracking under last month's pace overall, but two areas stand out. Building 2's conference rooms are running HVAC outside booked hours — which coincides with the May 14 all-buildings peak of 9.4 MWh. And water use at Greenhouse cluster B is up about 12% week-over-week.
See what matters before you search
Watchers and anomaly detectors run scheduled prompts through the model and present findings as cards. Pull any card straight into a conversation.
- Scheduled scenarios & watchers
- Severity-ranked, auto-resolving findings
- One tap to investigate in chat
Insights
6 active · 11 resolved todayLive findings from watchers, anomaly detectors and scheduled scenarios. Click any card to pull it into a conversation.
Climate units in rooms 4-201 to 4-208 have been running since 06:00 — three hours before the first booked meeting. Estimated waste this morning: 42 kWh (~€8).
Pull into chatMean indoor temperature across 6 zones in the North wing is consistently above the comfort band. HVAC schedule unchanged — pattern suggests a stuck damper or rooftop air handler.
Pull into chatRoom is booked frequently but actual occupancy stays low — typical "ghost meeting" pattern. Consider shrinking capacity in the booking tool or repurposing as a quiet zone.
Pull into chatAir moisture is climbing despite stable outdoor weather. The current irrigation cycle could likely be shortened by 12–18 minutes without affecting target soil moisture.
Pull into chatLights stay on past 23:00 three nights this week with no occupancy detected. Suggest tightening the auto-off window or adding a motion override.
Pull into chatFlow was at 2.4× baseline for ~50 minutes. Coincides with the weekly deep-clean window — pattern matches the last three Mondays. No leak indicated — worth confirming against the cleaning schedule.
Pull into chatSave your go-to scenarios and run them in one click
A workflow is a preset of prompts the assistant runs in sequence — save the questions you ask most often once, then rerun the whole flow whenever you need it.
- Multi-step prompt presets, run top to bottom
- Save your most-used scenarios once
- One click to rerun an entire sequence
Scheduled checks that watch the system for you
A watcher is a saved prompt sequence the assistant runs on a schedule — so you get a status report on time, or a check fires and notifies you the moment something drifts.
- Prompt sequences run by the LLM on a cron
- Scheduled status reports & health checks
- Notifies you only when something needs attention
LLM reasoning over a live GraphQL layer
PixelMonitor classifies and models the data upfront, so the assistant queries the same objects, properties and history as every dashboard and rule — which is what makes an answer checkable.
Your question
Plain language, any team
Assistant Agent
LLM plans the query and structures the answer
GraphQL layer
Live objects, properties and history
PixelCore
PostgreSQL · TimescaleDB
What PixelAssistant does not do
Stating the limits is part of making an answer usable. An operator should know exactly how much weight to give it.
It surfaces possible contributing factors and the evidence behind them. Confirming what actually caused a condition remains an engineering judgement.
The assistant does not change set points or run plant. Anything that acts on the estate goes through configured workflows and the people responsible.
Where a projection is shown, its historical basis is stated. It is not a validated predictive-maintenance model for an asset class.
Answers cover data that is connected and modelled in PixelCore. Anything outside the model is outside the answer — and the assistant says so.