No jargon. No acronyms you need a PhD to understand. Just a clear explanation of what the platform does, why it matters, and what you can expect if you deploy it.
If you'd prefer the technical version, see How It Works →
It decides when to charge and discharge a battery storage asset — and it makes that decision better than a human or a simple rule-based system can.
If you own a large battery that sells electricity into a wholesale market, the difference between a smart dispatch decision and a dumb one is real money every day. Avrenix's nine engines work together to make the smart decision every five minutes, around the clock, automatically — and a pilot proves the difference on your own data before you commit.
If you're a commercial building or industrial facility with large electricity bills, it works out how to stop you being charged for demand peaks — cutting the demand charge, proven on your facility's own interval history before you act on a single recommendation.
Because every dimension of the problem requires a different kind of intelligence — and those intelligences need to talk to each other.
Making a good dispatch decision requires knowing: what the price will be in 4 hours (forecasting), what regime the grid is in (classification), what happens if the forecast is wrong (adversarial planning), how your own dispatch moves the market (microstructure), and whether you can earn more from ancillary services than spot arbitrage (portfolio optimisation). No single model handles all five.
Existing platforms typically handle one or two of these dimensions and approximate the rest. Avrenix handles all five simultaneously, with each engine's output conditioning the next.
Not with a number from our marketing deck — with a test on your own data. Before you commit, Avrenix replays its dispatch strategy against your site's own price history and compares it to the alternatives, including the fixed "charge overnight, discharge at peak" schedule most operators run today.
The test is done properly: walk-forward — the strategy is trained on past data and scored on later, unseen data, exactly as it would run in production — against several standard baselines, with a statistical significance test (Diebold-Mariano) to confirm any difference is real and not luck. You get the working, not just the answer.
And the answer isn't a single figure. It's a range of outcomes — because markets are uncertain and an honest projection says so.
Avrenix doesn't promise a specific dollar figure. It gives you the bad-, likely- and good-case outcome on your own data — which is the honest answer, and the one a lender or board can underwrite.
Three things, honestly stated:
1. The forecast is wrong. Avrenix's forecast is better than most alternatives, but it's still a forecast. In unexpected grid stress events — major outages, weather extremes — the forecast degrades for 4–8 hours before recalibrating. The adversarial planning (FENRIR) partially protects against this, but not completely.
2. The data connection goes down. If your BMS or price feed disconnects, Avrenix switches to a conservative fallback mode — it recommends holding at the last known state rather than acting on stale data. It won't push you into expensive mistakes on stale data, but it also won't be optimising actively.
3. Market structure changes. If the ISO significantly changes its market rules or settlement methodology, the models need to be recalibrated. This has happened historically (e.g. ERCOT's settlement changes post-Uri). When HELIOS detects persistent drift from its predictions, it flags this automatically and triggers retraining.
None of these risks are unique to Avrenix — they apply to any dispatch system. What Avrenix adds is visibility into when each of these is occurring, rather than discovering it after the fact in your settlement statements.
Take a mid-sized data centre or manufacturing facility. You might have a monthly electricity bill of $800,000 — of which $250,000 is the demand charge (a fee based on your single highest 15-minute peak usage of the month).
The demand charge is disproportionately painful because one bad 15-minute period at 3pm on a hot Tuesday determines a month of charges. If your peak was 1,400 kW but PHALANX discharged your battery to limit it to 1,000 kW, you've saved 400 kW × your demand rate. At $18/kW/month, that's $7,200 saved in one intervention.
That is one intervention. Across a full year, consistent peak management plus time-of-use optimisation and export revenue add up — and the exact figure is what a pilot measures on your facility's own interval history, before you act on a single recommendation. We quote your number, not a brochure number.
Chai is the conversational interface for the platform. Instead of reading a dashboard and trying to interpret what the engines are telling you, you ask a question in plain English and Chai answers it.
"Why did CONSUL counter-bid this morning?" — Chai explains the opportunity cost calculation. "What's the probability we miss the demand threshold this month?" — Chai pulls the PHALANX risk assessment and tells you in one sentence. "How much revenue did we lose because of the forecast error on Tuesday?" — KAIROS's causal analysis, translated into dollars.
For operators who are technically sophisticated, the cockpit dashboards give full visibility. For executives who need a brief, Chai is the interface. It answers questions you'd otherwise have to schedule a meeting to ask.