Nx·Studio is where your process owner scopes a digital worker and your team sets its rules and approvers. Nx IDE lets your engineers deliver the changes your teams request, with editor actions approved first by default. The Cognitive Engine takes routine cases off your reviewers' queue only after its rules pass an exam.
Your process owner documents today's process and names the target outcome. The team then agrees which routine volume, delays and exceptions the worker should take on, and checks the capacity and cost assumptions.
Your team watches the worker handle generated test cases while outside systems are simulated. Each trial run reports accuracy, cost and safety scores, and no live record is touched.
After handover, your operations and IT teams run the worker on your own systems. Nxgyle charges no usage fees after export, and your model vendor bills AI use directly.
Logic your team can explain, approval records your auditors can check and rules that behave identically on every run. Our focus is finance, healthcare and other regulated operations.
Your engineers choose among 31 built-in modes, 21 of them specialists. If the Orchestrator leads a change, separate agents test and review it without seeing the author's reasoning, and a person on your team decides whether it ships.
Team leads match a model to planning, coding or review from 20+ supported vendors, including local models that keep code on your own machines.
Editor actions wait for approval by default. Allow, ask and deny rules set the boundaries, hooks run your checks, optional cost caps limit spend and checkpoints restore in one click.
Cortex builds a code map on each workstation, offline, and shows which parts a proposed change could affect before anyone approves it.
Reviewers see every approval checkpoint, threshold and hand-off point, with the approver for each, on a canvas or in swimlanes, ahead of live operation.
Each approval is kept on record. Two optional add-ons go further: field-level lineage, tracing each output to its input, and a tamper-evident run log. Auditors can request the audit-trail specification.
Every export carries a signed manifest of its software and AI components for engineers to verify offline, plus a threat model of its security risks.
Trials can run on generated test data alone. Scoping works from your process documents, so no production data has to leave your systems to begin.
Each step has an owner on your side and evidence the team checks before the next step starts.
The owner sets out today's process, as a short narrative or through the guided 10-step form, and attaches the team's procedures. Optional modules also accept BPMN diagrams, process-mining logs and the code of agents already in use. Data-processing terms are agreed before anything is shared, and what you provide is used only to design the worker.
Team checkpoint: the scope is agreed in writing, and trials can stay on generated test data from first run to last.
Nx·Studio's AI business analyst prepares several designs (five at most), each pitched differently from conservative to cost-optimal, then ranks them and puts one forward. The process owner chooses, and your team sets the approvers and the rules each step must respect. Connections draw on a catalog listing thousands of APIs; your engineers complete the starter code for Salesforce, SAP, ServiceNow, Workday and Oracle EBS against your own instances. Any system that exposes a standard web API, or an MCP server (an open standard for AI-to-system connections), can be connected too.
Team checkpoint: reviewers read the chosen design's full blueprint, steps and approvers included, before the first trial runs.
Nx·Studio generates test cases for the worker's own process and simulates every outside system. Your team runs the whole flow or one step at a time, and approvers answer the same pauses they will meet in live operation. Several workers can trial together on compressed time, as simplified stand-ins by default or as the real workers. Optional red-team testing tries known attacks against them.
Team checkpoint: each trial run is scored on accuracy, safety and cost, and two or three versions can be compared on one scenario before anyone approves.
When approvers are satisfied with the trials, IT receives the full project as a ZIP. It deploys the worker on infrastructure you run, with the models you select, local ones included. Operations then run the worker day to day: no vendor runtime, no Nxgyle usage charge and no call home.
What each export package holdsTeam checkpoint: IT checks the signed component list offline (every model, agent and software dependency is named) and compares the ownership report with what arrived.
Process owners scope and trial each worker in Nx·Studio, the flagship. Engineers deliver changes in Nx IDE, inside VS Code. A technical team connects the Cognitive Engine to repeat decisions, and reviewers approve workers against the Foundation’s checks. Versions and approvals stay on record at every handover.
Your process owner brings one workflow and its rules. Nx·Studio proposes costed designs, five at most, and recommends one; the team sets approvers, trials it on generated cases, then IT takes delivery of the code.
Engineers work each change request through specialist roles inside VS Code, from the plan and the code to an independent test and review. Your team picks the model vendor or a local model; the editor asks for approval by default.
Your reviewers confirm outcomes, and the engine turns them into rules. A rule decides only after a perfect score on unseen cases; until then your team decides. It is reached through a separate API, and published results are synthetic.
Every Nx·Studio worker carries four controls, so managers and reviewers check the same evidence before it goes live.
Model quality is converging, so a rollout succeeds or fails on how the worker is controlled. Each Nx·Studio worker carries the Foundation's four mechanisms, and every one leaves evidence that operations, IT and risk staff can check for themselves.
Nx·Studio builds realistic test data for the specific process a worker covers and stands in for outside systems. Your team runs trials step by step or end to end, and each run stops at every approval checkpoint until an approver responds. Joint trials of several workers run on compressed time. Every trial earns accuracy, safety and cost scores; no production data is needed.
Thresholds and limits from your procedures become fixed rules inside the worker wherever the AI analyst can encode them. The process owner accepts each rule, and anything that could not be encoded is listed for the team to resolve. A fixed rule settles identical inputs identically and never calls an AI model. Routine judgments that live only in people's heads are better suited to the Cognitive Engine: its own product, connected through its own Studio and learning from confirmed outcomes.
A readable blueprint, not a pile of prompts, sets out each worker. It names every step, each approver, the hand-off points and the systems involved. Every save creates a version that your team may compare with others, branch or merge under change control. Managers follow the same design as swimlanes, one lane per team, person or system. At export, an itemized list shows the delivered contents, and the blueprint remains your organization's property.
The export screen lists the full package before anything downloads. Its signed, tamper-evident AI bill of materials names each software and AI part, and your engineers verify it offline using the verifier shipped alongside. A security threat model comes with it, along with EU AI Act documentation in template form. After handover the worker runs without Nxgyle: no runtime from us, no license check, no metering and no calls home.
Your approvers keep the final say: at each approval checkpoint the worker waits for an authorized person. Exports that contain checkpoints also carry a reviewer inbox application; your engineers hook it up to the worker. Approvals in Slack are optional. Team leads can also set confidence thresholds: by default a low score raises a warning, and such cases can be routed to a person.
Teams that run AI agents often pay for a model call on every decision, routine or not. The Cognitive Engine works beside those agents and takes over the routine share. It bases its rules only on results your systems or reviewers have confirmed, and no rule may act before it has cleared an exam. Qualified rules answer in milliseconds without a model call, and your team can read every one in plain words.
Your engineers connect it via the decision API that the Cognitive Studio provides, and every result so far comes from simulated companies running synthetic data.
Your reviewers' confirmations count as evidence; an AI's proposals never do.
The default exam: a perfect score on 400 or more cases kept out of training.
No model call per decision, and a given release gives the same answer to the same facts.
Cases it cannot judge land in your reviewers' queue; an AI model is consulted only with your consent.
A blueprint sits behind every worker: a versioned design holding its roles, connections, approval checkpoints, routing and rules. When the team edits one control, reviewers see that single change spelled out in plain words.
Illustrative: one approval rule changed between two versions. Process owners follow the design as swimlanes by team, person and system. Engineers compare and merge versions as they would code, and auditors receive a plain account of the build. Nx·Studio validates the blueprint each time it is saved and exports it as source code for the environment you choose, owned by your organization.
First, sort the steps that need AI judgment from those fixed rules can settle. Stated rules turn into fixed checks where the analyst can encode them, and a fixed check never calls a model. Before anything is built, the AI analyst gives each design option a monthly run-cost estimate, and the Studio flags AI steps that a rule could replace.
The remaining AI work runs on model accounts your organization contracts, fully local models included. Team leads can set optional limits on the steps a run may take and how long it may last. Where agents face the same decisions daily, qualified routine cases can be handed to the Cognitive Engine; it decides them with no model call. That engine is sold as its own product, with its own Studio as the point of connection.
Your team runs the exported code, so there is no runtime license and no Nxgyle meter on decisions. The model vendor bills the AI calls your workers make directly to you.
Plan two budget lines: the scoped design and delivery fee Nxgyle charges, and the running costs your team manages for infrastructure, AI usage and maintenance. Optional support is scoped separately.
Fixed checks settle these steps with no AI model involved, so this part of the bill stays at zero.
Each design option carries the AI analyst's estimate of hours saved and monthly running cost; the generation plan adds the payback estimate. Trial runs then report each run's actual cost.
Your team runs the exported code, with no runtime license and no Nxgyle meter that rises as decisions grow.
Work through them before the first process goes live. Each one tests how a worker would run inside your team's routines.
Before the worker acts, can your reviewers read its logic, or only afterward?
If the vendor contract ends, which workers keep running?
Does your team receive every line of source code?
Will a busy month raise the vendor's fees, case by case?
Who notices, and what happens, if the model's behavior drifts?
Can a crafted input talk the worker out of its policy?
Is the worker your team runs the same one it trialed?
Can trials use realistic cases with every live system disconnected?
Can the evaluation start without your production data?
Are your approvers built into the worker's code, or held in a setting?
Can IT confirm, on its own hardware, that nothing depends on the vendor?
Is each number you are shown backed by a stated way of measuring it?
Each one is answered on this page.
Nxgyle builds digital workers your organization owns and runs. On this site, three words always mean the same thing:
Software your team owns, built for one defined process, which follows your rules and stops at your approval checkpoints.
The worker's readable, versioned design; reviewers check it before every change goes live.
The source-code project IT receives, then configures, tests and operates on your own infrastructure.
Sample export: files, README, testsWhite-label business applications, generated and delivered complete.
Each runs one defined process under fixed rules, and its source code is yours.
Catalog coverage spans finance, healthcare, manufacturing, energy, legal, logistics and automotive.
Each count is calculated from the portal and worker records listed on this site. A digital worker is an operational asset in code: it runs one defined process under your rules and approval checkpoints, and your organization holds the full source.
Nxgyle runs design and build while your process owners review each stage, and can remain engaged for upgrades; everything produced is yours outright.
Nxgyle engineers extend your agent systems and the applications built on them, working to your team's roadmap and priorities.
Senior AI specialists and engineers join your teams and work under your managers, adding delivery capacity where you need it.
