Skip to content
Ryedore Logo
Theme
ContactRequest Demo
Ryekronix operating console in an industrial control room

Ryekronix · agentic operations assistant

The agentic AI for industrial operations. Ask. It answers, plans — and proves.

Ask in plain English. Ryekronix plans the investigation, runs the right tools on your live data, checks the answer against physics and standards, and asks a human before any action — for any plant with operational data: chemical, oil & gas, utilities, manufacturing.

Watch a full runRequest a live demo
35+ tools
Tools it plans across
attested as of Aug 2026
6 checks
Checks before any action
attested
100% on‑premises
Where it runs
attested
in one picture

What Ryekronix does, in six words

Ask · Sees · Thinks · Proves · Acts (gated) · Learns. That is the whole product. Everything below is one of these six, with an example.

Ryekronixoperations agentAskYou ask in plain English“Why is R‑201 trending toward an alarm?”SeesYour live data + knowledge base + standards24 h of streams, the operating envelope, the SOPThinksPlan → execute tools → observe → revise35 tools; changes plan if evidence demandsProvesEvery step visible, with a confidence91 % match to a prior incident · confidence 87 %Acts — gatedSix checks, then a human approvesWork order or retrain — never without a personLearnsGets better from its own tracesTools that led to confirmed answers rank higher
Autonomous in thinking, gated in acting: the two green branches are the only places Ryekronix touches the plant — and never without a person.

swipe → to see the whole diagram

watch · under 4 minutes

Industrial AI that thinks before it acts

How Ryekronix takes a plain‑English question through plan, evidence, validation and a human approval.

Not a chatbot. An operations agent.

It plans, runs real tools on your live data, and shows every step — the table below is the difference, row by row.

the same question, two kinds of answer

A chat assistant versus Ryekronix — on one real question

Ask both: “Why is Reactor R‑201 trending toward an alarm, and what should I do?” The rows are what each one actually does. illustrative

What
A chat assistant
Ryekronix
How it starts
Reads your sentence and writes a plausible paragraph from what it was trained on.
Turns the question into a plan: which data to pull, what to compare it against, which checks to run.
on R‑201 · Pull 24 h of R‑201 streams → compare to the operating envelope → cross‑reference similar reactors → validate against physics and regulatory bounds.
What it looks at
Nothing of yours — it has no access to the plant.
Your live data, your knowledge base, the relevant standards — on your hardware.
on R‑201 · Jacket ΔT +3.2 °C from setpoint; catalyst‑bed pressure −0.4 bar in 6 h and accelerating.
How it reasons
Pattern‑matches text; cannot tell you where a claim came from.
Reasons in the open: observation → deduction → analogy → cause chain, each step shown with its evidence.
on R‑201 · 91 % match to a March‑2024 catalyst‑poisoning incident; chain: feed impurity → fouling → channeling → hot spot.
How sure it is
Sounds equally confident about everything.
Gives a calibrated confidence and escalates to a person when it is unsure.
on R‑201 · Thermal model confirms 14–18 % activity loss; confidence 87 %.
What it can do about it
Nothing — it can only write.
Proposes one of two actions (a work order, a model retrain); six checks and a human approval come first; refusals are logged.
on R‑201 · Reduce feed to 75 % within 4 h; schedule catalyst sampling — the work‑order request waits in Approvals.
What you can check afterwards
A transcript.
The full trace: inputs, tools, evidence, confidence, the decision — every step a row you can audit.
on R‑201 · Open the run: every tool call and every number it used, in order.

The chat assistant is not wrong to exist — it is the wrong tool for a plant. Ryekronix is built to be checked: every row above leaves a trace you can open.

one question, end to end

One question. The full run.

Six steps from a plain‑English question to an approved action. illustrative

QuestionPlanObserveReasonValidateRecommend → approval
  1. 1 · Question

    Why is Reactor R‑201 trending toward an alarm, and what should I do about it?

  2. 2 · Plan

    Pull 24 h of R‑201 streams → compare to the operating envelope → cross‑reference similar reactors → validate against physics and regulatory bounds → recommend.

  3. 3 · Observe

    Jacket ΔT +3.2 °C from setpoint. Catalyst‑bed pressure −0.4 bar in 6 h, and accelerating (0.05 → 0.08 bar/h).

  4. 4 · Reason

    91 % match to a March‑2024 catalyst‑poisoning incident. Chain: feed impurity → catalyst fouling → bed channeling → hot spot.

  5. 5 · Validate

    Thermal model confirms 14–18 % activity loss; safety envelope crossed in ~9 h at this rate. Confidence 87 %.

  6. 6 · Recommend → approval

    Reduce feed to 75 % within 4 h; schedule catalyst sampling. The work‑order request waits for a human in Approvals.

what its findings look like

A finding, its evidence, and one recommended action

Captured from the platform on demonstration data.

Product screen: fleet intelligence summary — an ensemble finding with causal and knowledge-base tags and a recommended action
The whole fleet in one finding — tagged, sourced, actionable · Screenshot from the product — demonstration data.
autonomous in thinking · gated in acting

Every action passes six checks and a human — and refusals are logged

Ryekronix can propose only two actions: dispatch a work order, or trigger a model retrain. Neither happens without six checks and a human approval. Refusals are logged too.

AUTONOMOUS IN THINKING · GATED IN ACTINGRyekronixproposes an action1role2policy3payload4physics limits5blast radius6duplicatehuman approvesthen it actsREFUSED at any check — and loggednothing is dispatched; a person decides what happens nexttwo action types only: dispatch a work order · trigger a model retrain
Only two action types exist (dispatch a work order, trigger a retrain). Every request, approval and refusal is a row you can audit.

swipe → to see the whole diagram

A robot cell held under a holographic check
when a check fails

It says STOP — and tells you why

When a proposed action fails a check — a setpoint outside SOP bounds, a physics limit, a duplicate — Ryekronix refuses, shows the reason, and logs it.

Nothing is dispatched. A person decides what happens next.

the suite

Six surfaces, one agent

Console, Approvals, Design Studio, Quick What‑if, Objectives, Learning — everything that plans and can act is Ryekronix.

Each one below with what it does and one example.

An industrial site at night, seen from above — the operations Ryekronix watches
the suite

Each surface, with one example

  1. 1 · Console

    Ask in plain English; get the answer, the plan and the trace.

    for example · The R‑201 run, end to end, in one screen.

  2. 2 · Approvals

    Every proposed action waits here for a person — after six checks.

    for example · A setpoint outside SOP bounds is refused by the physics check and logged.

  3. 3 · Design Studio

    Describe an equipment train; Ryekronix composes it and previews the KPI effect.

    for example · “A 2‑stage compression train with intercooler and aftercooler” → linked components, boundary checks, a diagram.

  4. 4 · Quick What‑if

    Rank setpoint moves in seconds — it proposes, it never dispatches.

    for example · Bearing temperature high → five candidate setpoints ranked by gap‑closure vs risk. A fast estimate, not physics — that is the Lab.

  5. 5 · Objectives

    Set the targets it reasons toward.

    for example · Close the throughput gap on K‑101 without exceeding vibration limits.

  6. 6 · Learning

    It gets better from its own traces, nightly.

    for example · Tools that led to confirmed answers rank higher next time.

Ryekronix vs the Specialists Lab

Which one, when

Something is wrong right now? Ryekronix. Considering a change? The Specialists Lab tests it on a virtual copy of the machine first. Ryekronix can commission the Lab; the Lab never touches the plant.

Meet the Specialists Lab →
Diagnose & act
Prove before budget
Wrong now
Ryekronix
Investigates, explains, proposes an action for approval.
Can commission the Lab for a deeper check
Considering a change
Ask Ryekronix — it points you to the Lab
Specialists Lab
Six specialists test the change on a virtual copy; statistics back.
what the round builds here

Where Ryekronix goes next

The one program that belongs to the agent — stated as today, next and the proof we will sign before we call it shipped. roadmap · not shipped

program 6

A sovereign reasoning model — and causes it can compute, not just cite

today · measured
Ryekronix plans across thirty‑five tools, passes six checks and a human before any action, and learns nightly from its own traces; every answer carries a cause chain from its causal reasoning.
next
An on‑premises reasoning model distilled for industrial operations from the knowledge bases, physics rules and traces — no external model dependency; offline learning from approved‑action traces; formal plan verification ahead of the gate; do‑calculus so ‘what would have happened if…’ is a first‑class answer.
proof point
Answer and refusal precision on held‑out operator questions; zero egress verified; cause‑chain precision against confirmed root causes.

Funded by the growth round — what the round builds, all ten programs →

questions people ask

Ryekronix, in five answers

Is Ryekronix a chatbot?
No. A chat assistant matches a pattern and returns text. Ryekronix plans a task, runs real tools on your live plant data, observes what came back, revises if the evidence demands, and shows every step with a confidence.
Can it act on the plant by itself?
No. It can propose only two actions — dispatch a work order or trigger a model retrain — and neither happens without six checks (role, policy, payload, physics limits, blast radius, duplicate) and a human approval. Refusals are logged.
Where does it run and what data does it see?
On your hardware, inside your network. It reads your live operational data, your knowledge base and the relevant standards; nothing leaves the site.
Does it depend on an outside model?
The platform’s predictions come from our own industrial multi‑task encoder — a shared cross‑industry representation, no third‑party model anywhere in the stack. Ryekronix’s planning runs on‑premises against your data, knowledge base and standards; an on‑premises reasoning model distilled for industrial operations is on the roadmap and labelled as such until its proof point is signed.
Which industries does it work in?
Any industry with operational data — chemical processing, oil & gas, utilities, manufacturing, mining, pharmaceuticals, healthcare equipment and more. Aerospace, defence and nuclear run at a manual‑approval tier floor.
where next

Ryekronix touches every page

See Ryekronix work on a problem from your operations.

A live run on your data or ours — plan, trace, validation and the approval step.

Request a live demoHow the platform reasons