Core modules
// LANE-2 · the three lanesFATHOM Forecast
Probabilistic generation, demand and price forecasts from sub-second to D+7. Calibrated quantile bands. Edge or cloud. The model that feeds everything else.
→ open moduleFATHOM Dispatch
Stacked-revenue optimisation across wholesale, balancing, ancillary and capacity markets. MISP with graph-attention warm-start. P99 solve under 1.2 s.
→ open moduleFATHOM Sentinel
Physics-informed anomaly detection on transformers, switchgear, inverters and BESS modules. Flags incipient faults weeks before threshold alarms.
→ open moduleConnectors
// LANE-3 · the boring layer that mattersThe connector tier is the unglamorous, mission-critical layer that takes the bulk of any real deployment. We ship six first-class connectors plus a generic protocol toolkit. None of them are afterthoughts; each one has its own conformance test suite that is re-run on every release.
Runtime & supporting services
// LANE-4 · what the modules run onFATHOM-RT (runtime)
Single-binary Rust runtime hosting model inference, connector lifecycle, and observability. Runs on edge (substation), regional VM, or operator workstation. ONNX Runtime + TensorRT inference acceleration.
version 0.9 · GA Q1 2026FATHOM-TS (time-series)
Columnar time-series store with native 100 ms resolution, IEC 61850 logical-node indexing, and replay-from-cold-start capability. Built on Apache Arrow + Parquet, with a Postgres-compatible query head.
version 1.2 · GAFATHOM-OBS (observability)
OpenTelemetry-native traces, metrics and logs, with a domain-specific span schema for dispatch decisions: forecast inputs → optimiser run → set-points emitted → actuator confirmation. Audit-ready.
version 0.7 · GA Q2 2026FATHOM-CTL (control plane)
Tenant, identity and policy management. OIDC integration, hardware-token enforcement for write paths, role-based dispatch envelope limits, configuration as code with full git-backed audit history.
version 0.8 · GA Q1 2026Licensing & commercial shape
// LANE-5 · how to buyWe license per asset-MW under management, not per seat and not per call. The reasoning is simple: a dispatch decision on a 50 MW battery is worth roughly fifty times a dispatch decision on a 1 MW battery, and pretending otherwise produces perverse incentives on both sides. Per-MW pricing scales with the value you extract; it does not penalise you for adding users or for running the optimiser more often.
The standard commercial path is a four-week paid scoping, an eight-week shadow-mode pilot (you pay for the data plumbing, we do not charge for the model output), and then a twelve-month minimum commitment on the assets we go live against. We have no minimum portfolio size — we have customers running 12 MW and customers running 480 MW on the same code paths.
For DNO and TSO engagements, the commercial structure is different: it tends to be regulatory-cycle aligned, with deliverables baked into RIIO-ED2 and Network Innovation Allowance projects. Talk to us about how that works in practice.
Pick a module, or just ask which one solves your problem.
If you are not sure whether Forecast, Dispatch or Sentinel is the right entry point, send us a one-paragraph description of your portfolio and the problem in front of you. We will tell you straight.