Skip to content
CHINOOKRequest early access

We build the verification layer for liquid-cooled AI halls

CHINOOK is software that checks whether a cooling plant's data is true and whether the plant does what its design says it will. This page explains why we started, what we've built, the rules we hold ourselves to, and, plainly, where we are today.

Why CHINOOK exists

AI training and inference are pushing rack power past what air can remove. Data halls are moving to liquid cooling: coolant distribution units, technology cooling loops and heat exchangers sitting between the IT load and the facility plant. Each of them adds sensors, meters and status points, often thousands per hall.

The way those halls are verified hasn't kept up. Commissioning still leans on spreadsheets, clipboards and stopwatches. After handover, the building management system shows green as long as each reading sits inside its alarm limits, whether or not the reading is true. A flow meter that reads 22 % low, or a temperature sensor frozen at one value for sixteen hours, can sit in a live hall for months without anyone noticing.

In a liquid-cooled hall, that matters more than it used to. Coolant margins are tighter, the equipment is less forgiving, and a wrong reading can drive a wrong control decision. We started CHINOOK to make the plant prove itself, with evidence a commissioning lead, an owner or an auditor can follow.

What we build

CHINOOK answers two questions every commissioning lead has to answer:

  • Is the plant telling the truth? Sixteen checks hold every sensor to signal quality, to physics (conservation of energy, the pump affinity laws, the system curve) and to its control signals. Every instrument gets a truth score.
  • Does the plant do what the sequence of operations says? Sequences become short test specs. On test day, each expectation is graded against live data as pass, fail or inconclusive, with the trend plot and the action log attached.

The platform is designed to be adopted in stages, each building on the last:

StageWhat a site gets
Trend auditUpload a historian export. CHINOOK maps the points, runs the checks and produces an evidence-backed report. No hardware on site.
Commissioning copilotSequences of operation extracted from project documents, test specs routed through approval, and an assistant whose every claim cites evidence.
Integrated systems testing and live dataTests graded against live data, read-only collectors and an edge agent that keeps working when the uplink drops.
Continuous commissioningThe checks rerun every 15 minutes against a baseline frozen at handover, and real events are graded against the handover tests.

The principles we build on

Read-only by default

CHINOOK reads your systems; it doesn't control them. That rule is enforced in the code, not just stated in a policy: inbound connectors only get a client that can read, and the protocol drivers contain no way to write. Writing setpoints is a separate component, off by default, that a site can only enable after independent assessment and named sign-offs.

Code decides, models explain

Deterministic checks decide what is true. Language models help draft explanations and answer questions, and every claim they make has to cite evidence. A confident wrong answer is worse than no answer in critical infrastructure.

Evidence travels with every finding

Each finding carries the rule and its version, the data window, the competing explanations with how well each fits, a plot, and a hash over all of it. Anyone can check how we reached a conclusion.

Unknown is not a pass

When data is missing, CHINOOK says so. A test with a gap in its data is graded inconclusive, never quietly passed.

Say what is verified, and what isn't

We describe what our software does and how far it has been tested, and we don't round up. That applies to this website as much as to a findings report.

Where we are today

CHINOOK is in early access. The platform described on this site is built and tested in our own environment, against real protocol servers (Modbus, OPC UA, BACnet Secure Connect, MQTT) and simulated plants with injected faults. The figures you see on this website come from that simulated site.

We are now working with our first design partners. To be specific about what that means:

  • We don't yet hold SOC 2 or ISO 27001 certification. Both are on our roadmap.
  • The component that can write setpoints has not been deployed at any site and will not be during early access.
  • We don't publish customer names without written permission.

If you commission or operate liquid-cooled halls and want to shape what we build next, join early access.

Who we work with

  • Commissioning leads and agents, who need defensible test records and sensors they can sign off on.
  • Data center operations teams, who want faults found before they cost uptime, and alerts that keep working when the network doesn't.
  • Owners, developers and MEP engineers, who need proof that a hall performs as designed.
  • Integrators and software partners, who can build connectors with our connector SDK and certify them against the same tests we use.

Get in touch

Write to chinookintelligence@gmail.com. We read every message. For specific requests, such as security reports, privacy requests or partnerships, see the contact page.