Decisions your AI has earned.
The Cognitive Engine learns how your business makes routine decisions, from results you have confirmed. It must pass an exam before it may act. Then it decides in milliseconds, without asking AI, with the same answer and a clear reason every time. Anything new goes to AI or to your people.
A separate product that runs on your own servers and works with the AI agents you already use.
Facts from your system
Rule, in plain words
A reading within its normal band, from a sensor that isn't stuck or dropping out, is OK.
Exam
1,453 of 1,453 right
Cases it had never seen
Independent checks
6 of 6 agree
Nx·Studio designs digital workers for your business processes. Nx IDE gives your developers an AI engineering team. The Cognitive Engine, a separate product, takes over routine decisions once it has proved it gets them right.
Asking AI for every decision doesn't scale.
Many AI agents ask a large AI model for every decision, including routine ones they have seen many times. That works in a demonstration. In daily operation it is slow, it is billed every time, and the answer can change from one day to the next.
- Seconds per decision Every answer waits for the AI model to reply.
- Pays for every decision Costs grow with every decision, including routine ones.
- Answers can change The same question can get a different answer tomorrow.
- Hard to audit An AI model's answer gives no reason your auditors can verify.
- Milliseconds A proven rule answers immediately, without waiting for an AI model.
- No AI call for routine work AI is reserved for new cases, within budgets you set.
- Same facts, same answer With the same rules, the answer never changes, and any decision can be replayed.
- A reason for every decision Each decision names its rule in plain words, and the exam it passed.
Think of a new trainee who has to earn trust.
No business lets a new hire approve decisions on day one. New staff learn from what actually happened and prove themselves on real cases. Only then do they work on their own, and they still ask for help when something is unfamiliar. The Cognitive Engine works the same way.
The employer
Knows what really happened. Its confirmed results are the only thing the engine learns from.
The tutor
Offers suggestions during training and, if you allow it, handles new cases. It never decides what counts as correct.
The trainee
Learns only from confirmed results, and must pass an exam before it works on its own.
The supervisors
Confirm results and settle unclear cases. In the strictest mode, they also handle anything the engine has not proved yet.
How the engine earns a decision.
Five steps, the same for every kind of decision. The engine takes over one kind of decision at a time, and only when the evidence says it's ready.
Learns from confirmed results
It studies past cases where your business has confirmed the right answer. AI can offer suggestions, but they never count as proof.
A factory confirms which past sensor readings were fine, and which needed action.
Passes an exam before it acts
Each rule is tested on past cases it has not seen. It may act only after getting at least 400 of them right, with no mistakes. You can raise this bar, but never lower it.
A rule that gets even 1 case in 515 wrong is not allowed to act. It must pass a new exam first.
Decides in milliseconds
A proven rule answers instantly, without asking AI, and the same facts always get the same answer. Six independent checks must agree. If they ever disagree, the rule is switched off.
A normal reading from a healthy sensor is marked OK, along with the rule that decided it.
Knows what it doesn't know
It never guesses. A case it has not learned goes to AI, within your budget, or to one of your people.
An unusual sensor reading is sent to AI or to a person, not guessed.
Keeps up as your business changes
Rules expire after 30 days, or sooner if you choose. A rule that turns out to be wrong is switched off at once, and must pass a new exam before it acts again.
If the factory changes what “normal” means, the old rule stops at the first case that proves it wrong.
Watch the engine work a shift.
Most of what an agent decides each day is routine. The engine takes over a routine decision only once it has proved it gets it right. Everything else goes to AI, within a budget you set, or to your people.
Nine sensor readings queued from the shift.
Proven rule
Passed its exam. Milliseconds, no AI call.
Not proven yet
Answered from your past decisions, or by AI within your budget.
A person decides
Sensitive or unclear cases go to your team.
Illustrative example: a factory where the engine has earned the routine sensor readings. Your mix depends on which decisions your engine has proved and the mode you choose. What never changes: proven rules make no AI call, AI runs within the budgets you set, and anything handed off reaches a person.
A routine decision the engine has proved makes no AI call at all. The AI calls that remain stay within budgets and daily limits you set, with an alert before a limit is reached.
How our pricing worksThe same facts and the same rules always give the same answer, even on days when an AI-only agent's answers change. Replay a shift, and every proven decision comes back identical.
Every engine decision names the rule that made it, in plain words. Six independent checks must agree on it. Every decision is recorded in a log that can't be edited, and anything handed off goes to a person.
Read the Trust CenterStart in shadow mode. Hand over once it has proved itself.
Choose how much the engine decides, for each application and each run. In shadow mode, its answers are scored against your system's own decisions, so you see how it performs before it acts.
Shadow
Your current system keeps making the decisions. The engine decides alongside it and is only scored, so you see how it would do before it affects anything.
Cognitive + AI
The engine handles what it has proved. Anything else is answered from your own past decisions where possible, then by AI within your budget, or by a person.
Cognitive
Only proven rules decide. Everything else goes to a person, or to your system's usual fallback.
A control room your business can read.
The Studio is where your team trains the engine, tests it, and signs off before it goes live. Each person gets the access their role needs.
Your business at a glance
See each application, its limits, and which decisions the engine has earned so far.
Side-by-side tests
Test the engine against an agent that asks AI for everything, or score it against your own system's decisions.
A reason for every rule
Read each rule in plain words, along with the exam it passed.
Your people stay in charge
Your team confirms results in a simple review queue, and approves anything AI drafts from your documents.
Go-live checklist and sign-off
A checklist for each application, and owner sign-off before anything goes live.
Controls and alerts
User roles, AI budgets and spending limits, an audit trail that can't be changed, and alerts when something needs attention.
Works with the agents you already have.
The engine sits beside your agents and systems rather than replacing them. Your systems stay in charge of what actually happens.
Connects to your agents
Your AI agents ask the engine for a decision through its API, each with its own access key.
Learns beside your system
Send it the decisions your system already makes. It decides alongside, without acting, and is scored against yours.
Developer tools
A Python toolkit for your developers, and an advice-only connection for AI assistants.
Runs on your servers
Installs on your own servers. After an upgrade, every rule must pass its exam again before it acts.
Safe with your tools. When the engine acts through your tools, it first records what it intends to do. It never retries an action on its own. If an outcome is uncertain, it stops and waits for your team.
Visit the Trust CenterWhat's included, what's optional.
Everything in the first list comes with the engine. Optional extras stay off until you switch them on.
Included
Ready to use- Learning from confirmed results, with an exam before any rule acts
- Go-live one decision at a time: proven rules act while the others wait
- Routine decisions with no AI call, cross-checked every time
- Hand-offs to your people, and rules that switch off when wrong or out of date
- The Cognitive Engine Studio, for training, testing and sign-off
- Connections for your agents and systems, installed on your own servers
Optional extras
Off unless switched on- Rules drafted by AI from your written policies. They must still pass the exam
- Rules that compare facts or look up related records
- Memory of past conversations, and learned preferences
- Extra safeguards for requests written in free text
- Automatic choice of the best method for each type of question
The questions buyers ask first.
Does it replace AI models?
No. It takes over the routine decisions it has proved it gets right. New cases still go to AI, within your budget, or to your people.
Will it ever act on a rule it hasn't proved?
No. Only rules that passed their exam may decide. Everything else goes to AI or to a person, depending on the mode you choose.
What happens when our business changes?
Rules expire after 30 days, or sooner if you choose. A rule that turns out to be wrong is switched off at once, and a replacement must pass a new exam.
Can we see why it made a decision?
Yes. Every decision names the rule behind it, in plain words, and you can see the exam that rule passed.
Does the AI model see our data?
Yes, when it is asked to help, with names and ID numbers hidden first. You can switch this off for any application or data source, and turn off drafting from your documents too.
How do we know it will work on our decisions?
Start in shadow mode. Your system keeps deciding while the engine decides alongside it and is scored against your answers. You see how it performs before it is allowed to act.