The Cognitive Engine starts empty and learns your routine decisions from past cases your people have confirmed. Before it may decide anything, it must pass an exam on cases it has never seen. Then it answers in milliseconds, the same way every time, and shows the rule it used. Anything new goes to a person, or to AI if you allow.
Facts from your system
Rule, in plain words
A reading in its normal range, from a sensor that isn't stuck or dropping out, is OK.
Exam
1,453 of 1,453 right
Cases it had never seen
Cross-checks
6 of 6 agree
From Factory Floor, a simulated test business with made-up data. Measured in October 2026.
Nx·Studio designs your digital workers. Nx IDE gives your developers an AI assistant. The Cognitive Engine takes over routine decisions once it has proved itself, so an AI model doesn't have to work out the same answer again and again. All three are built by Nexgile.
Many AI tools send every decision to an AI model, even ones they have answered many times before. That is fine in a demo. In daily work it is slow, each question costs money, and the answer can shift from one day to the next.
Every new case is compared with the rules the engine has proved. If one fits, the case is decided on the spot, with no AI model involved. If none fits, the engine does not guess: depending on the mode you choose, it goes to AI within your budget, or to a person.
New cases
Proven rules
Checked first
Decided by
Illustration only: the counts above come from this animation. Measured results are further down the page.
No business lets a new hire decide alone on day one. New staff study past cases, show they get them right, and only then work on their own. Even then, they ask for help with anything unfamiliar. The Cognitive Engine works the same way.
Owns the systems and knows what really happened. The engine learns only from outcomes your business has confirmed as right.
Offers suggestions while the engine learns and, if you allow it, handles cases the engine hasn't learned yet. It never decides what counts as right.
Starts knowing nothing about your business, learns only from confirmed outcomes, and must pass an exam before it works alone.
Confirm outcomes and settle cases where past answers disagree. In the strictest mode, they also handle anything the engine hasn't proved yet.
Every kind of decision goes through the same five steps. The engine takes over one kind at a time, and only when the evidence says it is ready.
It starts with no knowledge of your business. It studies past cases whose right answer your systems or your reviewers have confirmed. AI may suggest, but a suggestion never counts as proof.
The factory confirms which past sensor readings were really OK, and which needed action.
Each rule sits an exam of past cases it never saw while learning. It may act only after getting at least 400 right with no mistakes. You can raise this pass mark, but never lower it.
On a simulated factory with made-up data, one rule got 1 case in 515 wrong, so it was not allowed to act. With more examples, it later got 1,453 of 1,453 right.
A proven rule answers in a fraction of a second, without an AI model. The same facts always get the same answer. Six separate checks run the rule, and if they ever disagree, the rule is withdrawn.
A normal reading from a sensor that isn't stuck is marked OK, and the decision names the rule that made it.
It never squeezes an unfamiliar case into a rule. Depending on the mode, it reuses your past decision for exactly the same facts, asks AI within your budget, or hands the case to a person. Where past answers disagree, a person decides.
A reading unlike anything the engine has seen goes to AI or to a person, instead of being guessed.
Rules expire after 30 days, or sooner if you choose. The moment a confirmed outcome proves a rule wrong, it is withdrawn, and it must pass a new exam before it acts again.
If the factory changes what “normal” means, the first confirmed case that disagrees stops the old rule. Its replacement must pass its own exam.
Much of what an AI tool decides each day is routine. The engine takes over a routine decision only after proving it gets it right. Everything else is answered from your own past decisions, by AI within your budget, or by your people.
Nine sensor readings from the shift, waiting to be decided.
Passed its exam. Instant, no AI used.
Same facts as a past case? It reuses that answer. If not, AI helps, within your budget.
Handed off by design, or past answers disagree.
Illustration based on Factory Floor, a simulated business with made-up data, where the engine has proved itself on routine sensor readings. Your own mix depends on which decisions your engine has proved and the mode you choose. What stays the same: proven rules never ask AI, AI stays within your budget, and anything handed off reaches a person.
Proven routine decisions never use AI, so they add nothing to your AI bill. Any AI still needed stays within the budgets and optional daily limits you set, with a warning before a limit is reached.
How our pricing worksThe same facts always get the same answer, even on days when an AI-only tool's answers shift. Re-run a shift and every proven decision comes back exactly as before.
Every decision names its rule in plain words, and six separate checks must agree on it. Each one is kept in a record that can't be edited, and anything handed off goes to a person.
Read the Trust CenterYou set how much the engine decides, application by application and run by run. In Shadow mode it only practices: it decides next to your current system and is scored, but changes nothing.
Your current system keeps deciding. The engine answers alongside it and is only scored, so you see how it would have done before it changes anything.
Who decides, in order
The engine decides first. For a case it hasn't proved, it reuses your past decision for exactly the same situation. If your past decisions differ, a person decides. Only brand-new situations go to AI, within your budget.
Who decides, in order
Only proven rules decide. Everything else goes to the backup route your system already has, which is usually a person.
Who decides, in order
In the Studio your team gathers examples, trains and tests the engine, compares it with an AI-only tool, and approves it before it goes live. Each person gets the role they need, from viewer and reviewer to trainer and administrator.
Each application lists its AI tools, their limits, what must never go wrong, and which decisions the engine has earned so far.
On our simulated businesses, one click compares the engine with a tool that asks AI every time: accuracy, decisions left to people, speed and AI use. On your own system, it compares the engine with the decisions your system made.
See each learned rule in plain words, the exam it passed, and how each of the six checks answered.
Your people confirm cases from a simple list, including blind reviews that hide the suggested answer. The Studio tracks the quality of each review.
AI drafts each decision's facts, limits and policy from your documents and systems, citing a source for each. A person must approve every draft, and sources are re-checked weekly by default.
Before launch, each application gets a checklist, its owner's sign-off, a security review, and tests under heavy load and after failures. A hand-over pack comes with it.
Give each user a role and, if you like, sign in with your company login. Setting changes can need two people's approval, names and IDs are masked, and nobody can edit or delete the activity record.
Cap AI spending for each run, and add a daily cap per application if you like. You get alerts at 80% and 100%, and one switch can keep case facts hidden from AI entirely.
Every 15 seconds it looks for rules that are expiring or withdrawn, slow decisions, AI spending and low storage, and warns you.
We test the engine on twenty simulated businesses with made-up (synthetic) data, from insurance claims and payments to factories and energy markets. Each time, we compare it with an AI tool built the usual way, one that asks an AI model about every decision. Here is what we measured.
Field test on five simulated businesses with made-up data, October 6, 2026.
Every decision the engine made itself, on all 20 simulated businesses (made-up data, 1,000 events each, October 9, 2026). The engine currently decides cases in 14 of them.
Typical decision: 17–31 ms (thousandths of a second) for the engine vs 3.3–4.2 seconds for an AI-only tool. Five simulated businesses, made-up data, October 8, 2026.
Same simulated runs, made-up data: per 1,000 decisions, the AI-only tool made 329–1,519 AI requests; Cognitive + AI mode made 0–14. Most AI-only figures are scaled up from runs capped at 300 requests.
On the same simulated factory decision, an AI-only tool using the same AI model and the same instructions scored 97% one day and 20.5% the next. The engine's proven rule decided the same 865 readings on both days, and got none of them wrong.
All figures were measured by Nexgile in October 2026 on simulated businesses with made-up (synthetic) data. “Right” means matching each simulated business's own policy. They show the method, not a customer's results, and do not by themselves prove the engine is ready for live use.
Nothing gets ripped out. The engine works next to your existing AI tools and systems, and your systems stay in charge of what actually happens.
Any AI tool can request a decision over a standard web connection (an API), using its own access key.
Send in decisions your system has already made (a shadow feed). The engine answers alongside without acting and is scored against yours. New situations become cases for your people to confirm.
AI assistants can consult the engine through a local MCP server, the open standard AI tools use to reach other systems. Here it only advises and never acts. Developers also get a Python toolkit (SDK).
Arrives packaged to run on your own servers (with Docker Compose). After an upgrade, every rule must pass its checks again before it acts.
Careful with your systems. Before the engine acts through your tools, it records what it is about to do. It never retries a failed action on its own, and if a result is unclear, it stops and waits for your team.
Visit the Trust CenterEverything below is built and available today. The extras stay off until you switch them on.
No. It takes over only the routine decisions it has proved it gets right. New cases still go to an AI model within your budget, or to your people.
No. Only rules that passed their exam may decide. Depending on the mode you choose, everything else is answered from your own past decisions, by AI within your budget, or by a person.
Rules expire after 30 days, or sooner if you choose. A rule is withdrawn as soon as a confirmed outcome proves it wrong, and its replacement acts only after passing a new exam.
Yes. Every decision names its rule in plain words. In the Studio you can also see the exam that rule passed and how each of the six checks answered.
By default, yes: when AI is asked to help, it may see the case facts, with names and other identifying details from your data sources masked first. You can switch this off for each application or data source. AI also drafts from your documents unless you turn drafting off.
Not yet. Every result on this page was measured on simulated businesses with made-up data. In Shadow mode the engine decides alongside your own system without acting, so you can measure it on your own work before it acts.
