Models
Named, vertical reasoning models trained to reason over time-series data. Two sizes are available today, both trained on energy problems — starting with battery systems.
Why reasoning models
In industrial settings — energy, manufacturing, medtech — a good decision needs two things together: an understanding of what happened and why, and an estimate of what is likely to happen next. Today they live in separate tools. Root-cause analysis often starts and stops at a point detector, and forecasting runs in isolation from both.
Every silo has a price. What the diagnosis learns never reaches the forecast, and the physical limits of the asset never reach the diagnosis. A time-series reasoning model (TSRM) removes the seams: one line of reasoning from root cause, through event-based forecasting grounded in physical constraints, to an informed decision.
Today — three silos
With a TSRM — one arc
01 · The idea
A detector flags a deviation. A forecaster extends a curve. Neither tells you which cause is most likely, or what the asset can physically do next. Each handoff costs precision; a TSRM carries one line of reasoning from the past into the future.
02 · The evidence
The decisive clue is often not in the telemetry. Maintenance logs, event notices and operator notes are weighed with the signals in one step, so a diagnosis can draw on both, even for fault combinations no detector was trained on.
03 · The foundation
Text-token LLMs split numbers into subword tokens, lose arithmetic precision over long sequences, and get costly once raw telemetry is serialized as text. A TSRM works on the signal itself, so one arc stays precise and affordable.
RunnerTau 7B
Full time-series reasoning model — energy vertical
RunnerTau 7B is our full-scale time-series reasoning model (TSRM). It reasons about energy problems the way a domain expert would — understanding battery systems well enough to explain charge behavior, flag abnormal degradation, and reason about dispatch decisions against real operating constraints, not just pattern-match on historical curves.
Where an ordinary time-series foundation model forecasts a sequence, RunnerTau 7B reasons about why the sequence looks the way it does, and what constraints on the underlying asset make one continuation physically plausible and another not.
Domain
Energy — battery systems
Deployment
On-prem / private cloud
Fine-tuning
Optional, per application
RunnerTau 4B
Compact time-series reasoning model — energy vertical
RunnerTau 4B carries the same reasoning approach as the 7B model in a smaller footprint, built for constrained deployments — on-site hardware, edge gateways, or applications where latency and cost matter more than headroom.
It trades some capacity for efficiency, while staying grounded in the same asset knowledge and constraint-checking that defines the RunnerTau family. A natural fit where TEMPO needs a reasoning model running close to the asset itself.
Domain
Energy — battery systems
Deployment
Edge / on-prem, low-latency
Fine-tuning
Optional, per application
Roadmap
RunnerTau's reasoning starts in energy — batteries, generation, and grid assets — because that's where the time-series data and failure modes are best understood. The same reasoning approach extends naturally to three adjacent domains next.
Energy — battery systems (live)
RunnerTau 7B and 4B reason about energy problems today, including battery charge, dispatch, and degradation.
Data centres
Extending reasoning to power draw, cooling, and capacity planning for compute infrastructure.
Security & Defense
Extending reasoning to logs, alerts, and sensor telemetry — diagnosing incidents from text and numeric evidence together.
Automotive
From stationary energy assets to moving ones — powertrain, battery, and fleet telemetry.
RunnerTau is available for pilot deployments, optionally fine-tuned for your application.
Request a pilot →