The point isn't nine models — it's that they check each other. Every engine's output is immediately available to every other, so the system can explain why it acted and plan for the worst case, not just the average. That's what a black box can't do, and it's why the whole is sharper than any single model.
What this means for you: no single model gets the last word. Five explicit cross-engine links carry one engine's output into another's reasoning — the forecast feeds the risk model, the microstructure model constrains the dispatch — so a mistake in one place gets caught in another. These flows are inherent to the architecture, not configured per site.
SPECTER's learned grid-personality regimes shape FENRIR's robust plan and TARSIS's tariff optimisation — the same read of your grid drives both.
SYNDEX's Kyle-λ price-impact estimate becomes FENRIR's self-impact constraint, so the robust dispatch accounts for the market it will move.
HELIOS's distributional temperature forecast feeds PHALANX's peak-risk sampling — the demand-charge tail is drawn from a real forecast, not a point guess.
FENRIR's worst-case demand surge sets the peak TARSIS protects, so the tariff schedule is robust to the demand it can't perfectly predict.
CONSUL's ancillary negotiation and TARSIS's tariff schedule co-optimise over one shared capacity and state-of-charge budget with a correlated-activation model — the battery is never double-committed between market and facility.
Every engine reacts the moment something changes — no waiting on overnight batch jobs. Each engine publishes and subscribes to events with delivery guarantees across the platform.
Every engine sees the same picture of the world at the same moment — no two disagreeing on the facts. Grid-personality profiles, regime classifications, and forecasts live in one low-latency state store.
Seven FoM engines for operating a grid-connected battery in wholesale markets; two BTM engines for cutting a facility’s demand charges. Purpose-built, not repurposed — and wired together by the five links above.
Not a tenth engine: the intelligence layer over the nine. SHAP-style attribution on every dispatch, counterfactual narrative, and a complete audit trail with historical replay.
You see the numbers on your own history before a single live decision. Walk-forward validation against five baselines with Diebold-Mariano significance tests.
Built to run in production from day one, across a whole fleet. Multi-site dispatch with per-site state, connection pooling, rate limiting, and graceful load shedding.
Avrenix accounts are organizations, not single logins. Invite your operators, your analysts, and your executives — each with a role that decides exactly what they can do. The client hides what a role can't touch; the server enforces it on every request.
Owns the organization. Manages billing, members, sites, and engine runs — and configures what admins and members are allowed to do.
Runs the day-to-day: manages members and sites and runs every engine. Billing and org configuration stay with the Owner.
Operational access: manage sites and run the engines. No access to members or billing — the hands on the dispatch.
Sees everything, changes nothing. Built for executives and auditors who need the numbers without touching a control.
Add people by email with a single-use, expiring invitation and the role assigned at the door. Revoke any invite before it's accepted.
Tune what admins and members can do, within anchored bounds — a Viewer is always read-only, an Owner always has full control. Enforced server-side, every request.
Off by default. When an Owner turns it on, teammates from your own company domain can join automatically. Public email domains are refused.