The Cognitive Engine learns how your business decides, from outcomes you have confirmed. Before it may act, it must pass an exam on cases it has never seen. It then makes those routine decisions in milliseconds, with no AI call, giving the same answer every time and a reason you can read. Anything unfamiliar goes to AI or to your people.
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
Reasoning checks
6 of 6 agree
From the Factory Floor simulated business. Measured in October 2026 on synthetic data.
Nx·Studio designs digital workers for your business processes. Nx IDE gives your developers an AI engineering team. The Cognitive Engine takes over routine decisions that AI agents would otherwise re-think every time, once it has proved it gets them right. All three are built by Nexgile.
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.
Each case is checked against the rules the engine has proved. A match is decided in milliseconds, with no AI call, and the same facts always get the same answer. A case it hasn't proved is never guessed: depending on the mode you choose, it goes to an AI model within your budget, or to a person.
Work in
Proven rules
Checked first
Outcome
Illustration only: the counts above are from this animation. Measured results are further down this page.
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.
Owns the tools and systems, and is the authority on what actually happened. Its confirmed outcomes are the only evidence the engine learns from.
Offers suggestions while the engine is being trained. If you allow it, it also handles cases the engine has not yet learned. It never decides what counts as correct.
Learns only from confirmed outcomes, and must pass an exam before it may work on its own.
Confirm outcomes and decide cases where your past answers disagree. In the strictest mode, they also receive anything the engine has not yet proved.
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.
It studies past cases whose right answer your business has confirmed, either from your systems or by your reviewers. An AI model may make suggestions, but they never count as evidence.
The factory confirms which past sensor readings were in fact OK, and which needed action.
Each rule is tested on past cases it never saw while learning. It may act only after answering at least 400 correctly with no misses. You can raise this threshold, but never lower it.
On a simulated factory with synthetic data, one rule missed 1 case in 515, so it was not allowed to act. With more evidence, it later answered 1,453 of 1,453 correctly.
A proven rule answers in milliseconds, with no AI call. The same facts always get the same answer. Six separate reasoning tools run the rule, and all six must agree. If they ever disagree, the rule is withdrawn.
A normal reading from a healthy sensor is marked OK, and the decision names the rule that made it.
A case outside what it has learned is never forced into a rule. Depending on the mode, the engine reuses your own past decision for the exact same facts, asks an AI model within your budget, or passes the case to a person. Where your past answers disagree, a person decides.
A sensor reading outside every range the engine has seen is referred to AI or to a person rather than guessed.
Rules expire after 30 days, or sooner if you choose. A rule is withdrawn the moment a confirmed outcome proves it wrong, and must pass a new exam before it acts again. The engine changes only when the evidence does.
If the factory changes what “normal” means, the first confirmed case that disagrees stops the old rule. A replacement must pass its own exam.
Much of what an agent decides each day is routine. The engine takes over a routine decision only after it has proved it gets it right. Anything it hasn't proved yet goes to your own past decision or to AI within a budget you set. Anything handed off goes to your people.
Nine sensor readings queued from the shift.
Passed its exam. Milliseconds, no AI call.
Your own past decision for the same facts, or else AI within your budget.
Handed off, or your past answers disagree.
Illustrative scenario, based on the Factory Floor simulated business (synthetic data), 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 run within per-run budgets and optional daily caps you set, with an alert before a cap 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 separate reasoning tools 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 CenterChoose how much the engine decides, for each application and each run. On our simulated businesses, one click compares it with an agent that asks AI for everything.
Your existing system keeps deciding. The engine decides alongside it, and its answers are only scored. You see how it would have performed before it affects any outcome.
Who decides, in order
The engine decides first. For anything it hasn't proved yet, it reuses the decision your business made before in exactly the same situation. If your business decided that situation in different ways, a person decides. Only new situations go to AI, within your budget.
Who decides, in order
Only proven rules decide. Everything else goes to your system's safe fallback, which is usually a person.
Who decides, in order
The Studio is where your team gathers evidence, trains and tests the engine, compares it with an AI-only agent, and signs off before go-live. Viewers read the dashboards, reviewers answer questions, trainers run training, and administrators manage the rest.
Each application shows its agents, their hard limits, what must never go wrong, and which decisions the engine has earned so far.
On our simulated businesses, one click runs the engine against an agent that asks AI for every decision. It compares accuracy, decisions left to people, speed and AI calls. On your own system, it compares the engine with your system's decisions.
Read a learned rule in plain words, the exam it passed, and how each of the six reasoning tools answered.
Your people confirm cases in a simple queue, including blind reviews that show no suggested answer. The Studio tracks review quality.
AI drafts each decision's facts, limits and policy from your documents and systems, citing every source. A person must accept each draft. Cited passages are re-checked weekly by default.
A go-live checklist per application, owner sign-off, a security review, load and restore drills, and a hand-over package.
User roles, optional single sign-on, optional two-person approval for setting changes, masked identifiers, and an audit trail that can't be edited or deleted.
AI budgets per run and optional daily spending caps per application, with alerts at 80% and 100%. A switch decides whether AI may see case facts at all.
Warnings for expiring or withdrawn rules, slow decisions, AI spend and low disk space, checked every 15 seconds.
We test the engine on twenty simulated businesses with synthetic data, from insurance claims and payments to factory floors and energy markets. We compare it with an AI agent built the common way, which asks an AI model for every decision. Here is what we measured.
Field test on five simulated businesses with synthetic data, October 6, 2026.
Every decision the engine made itself, on all 20 simulated businesses (synthetic data, 1,000 events each, October 9, 2026). The engine currently decides cases in 14 of them.
Typical decision: 17–31 ms for the engine vs 3.3–4.2 s for an AI-only agent. Five simulated businesses, synthetic data, October 8, 2026.
Same simulated runs, synthetic data: per 1,000 decisions, the AI-only agent made 329–1,519 AI calls; Cognitive + AI mode made 0–14. Most AI-only figures are scaled up from runs capped at 300 calls.
On the same simulated factory decision, an AI-only agent 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 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.
The engine sits beside your agents and systems rather than replacing them. Your systems stay in charge of what actually happens.
Any agent system can ask the engine for a decision over the web, using its own access key.
Send decisions your system has already made. The engine decides 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 connect to systems). It advises and never acts. A Python SDK is also available.
Packaged with Docker Compose for your own servers. After an upgrade, every rule is re-examined 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 is built and available today, and the extras you can switch on.
No. It takes over 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. A replacement acts only after it passes a new exam.
Yes. Every decision the engine makes names its rule in plain words. The Transparency screen shows the exam that rule passed and how each reasoning tool answered.
By default, yes: AI may see case facts when it is asked to help, with identifiers and names 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 so far. Every result on this page was measured on simulated businesses with synthetic data. Shadow mode lets you measure the engine on your own system before it acts.
