Time-series reasoning models (TSRM)

The world's critical decisions
run on time-series.

Nearly every industry runs on time-series: sensors, telemetry, prices, logs, signals. Text-token LLMs were not built for it. We are building TSRMs, an emerging class of reasoning models that read time-series natively alongside text, documents and other modalities, to power private, agentic workflows, one industry at a time.

Reasoning AI for

medical data.

Patient Monitor 12

"Is this vitals drift consistent with the alarm history?"
Request a pilot → Not sure you're ready? Check your reasoning readiness →
Explore

The gap

Time-series is the hard part. Language models weren't built for it.

Text-token models read numbers as fragments of words. Across thousands of readings they lose precision, miss structure over time, and get expensive once raw telemetry is pasted in as text.

So most "AI for your sensor data" is a language model reading a summary, or a separate forecaster it never talks to. That is how decisions end up made in pieces. Why reasoning models →

Most tools also know nothing about the asset producing the data. They'll explain a vibration spike or extrapolate a forecast without checking whether the answer is physically possible for that specific machine. We start from your asset's own documentation — power curves, fault tables, system topology — so every explanation and forecast is checked against what the asset can actually do.

Not there yet?

Most data estates aren't reasoning-ready — telemetry, documentation, and asset context usually live in separate systems. Start with a Reasoning Readiness Assessment and find out exactly what's missing.

Check your readiness →

How it works

From a question to a grounded answer — placed back on your asset or process schema.

Underneath is a time-series reasoning model (TSRM). It reads telemetry natively alongside logs, manuals and notes, and carries one hypothesis from root cause to forecast to action. Why reasoning models →

01
Reads your asset's structure Power curves, fault tables, P&IDs, topology — ingested and mapped to your live telemetry channels.
02
Reasons over text and signals together Ask a question. The model reads telemetry and your maintenance logs in one step, selecting which signals matter, which analytical tools apply, and how your asset's constraints bear on the answer.
03
Finds it and shows you where It weighs the plausible causes, not just flags a deviation, then places the root-cause finding directly on your asset's own schema — a clogging filter, a degrading bearing — not buried in a log.
04
Checks before it tells you Forecasts and explanations are verified against real operating limits before they reach you.
05
Stays reliable when sensors fail When a sensor degrades, TEMPO falls back to related signals it understands — using the same asset structure to choose the right substitute.

Finding on asset schema

Heat pump schema with clogged filter finding System schematic showing a heat pump with outdoor unit, expansion valve, indoor unit, return line, circulation pump, and a water filter. The filter is highlighted in amber as the finding location, with a callout explaining the flow ratio has dropped 34% versus baseline, consistent with progressive clogging, and a maintenance log note that the filter was last serviced 400 days ago against a 365-day rated life. HEAT PUMP — UNIT 3 Outdoor unit Expansion valve Indoor unit RETURN LINE Circ. pump Filter ⚠ clogging Finding: water filter Flow ratio −34% vs baseline. Consistent with progressive clogging. Log: serviced 400 days ago; rated 365.

Offerings

Where we bring reasoning to your operation.

Reasoning readiness isn't limited to time-series. The same discipline — grounding an AI system in the structure it's reasoning about, then verifying every output against it — applies to your AI architecture, how you run inference, and your codebase. These are the ways we work with you today.

Offering

Reasoning Readiness

Is your operational data reasoning-ready? Our original assessment — for time-series, industrial data estates.

Explore →

Offering

AI Architecture

Deep expertise designing AI systems that reason over your data with precision, at scale, and with a full audit trail.

Explore →

Offering

Private Inference

The architecture and infrastructure to put AI to work locally, at a reasonable cost, fully under your control.

Explore →

Offering

AI Coding Adoption

From vibe coding to agentic engineering. We help you understand where you are and what it takes to reason over your codebase with precision.

Explore →

The agent & the models

An execution layer. A family of reasoning models. One ask.

TEMPO

Temporal reasoning agent

TEMPO reads your asset's documentation, selects the right analytical tools, and reasons over telemetry and asset constraints together. Ask a question — get an answer grounded in what your asset can actually do.

In development

RunnerTau

Time-series reasoning models (TSRM)

Named, vertical reasoning models trained to reason over time-series — not just forecast it. Today they understand energy problems, including battery systems, down to the level of a domain expert.

RunnerTau 7B AVAILABLE
RunnerTau 4B AVAILABLE
Explore RunnerTau →

Deployment

Put it to work. Locally.

TEMPO and RunnerTau run on your infrastructure, on your data, under your control. Nothing leaves your site unless you decide it should. They sit alongside the systems you already run — your historian, your SCADA, your digital twin — adding a reasoning layer rather than asking you to replace what's working.

◈

Data sovereignty

Your telemetry stays on-site. No cloud dependency, no third-party data exposure, no compliance risk.

◎

Works with your stack

Connects to your historian, SCADA, and digital twin investments. Adds reasoning on top — not a replacement.

◇

No lock-in

No dependency on a single model or vendor. Swap components as the landscape evolves.

Working on industrial time-series problems?

Design partner, early customer, or investor — we'd like to talk.

We respond to every message.