VEXPLOR
SERVICES — AI AGENTS ON YOUR DATA

Systems keep the records.AI agents make the decisions.

We do not replace the systems you use today. ERP, MES, PDM and even Excel — the systems and data your company already has are connected through an ontology, and an AI work team is built on top of them. Five agents that read data, forecast supply, propose orders, communicate with suppliers and write reports work together as one workflow.

Plug-in with no change to existing systems · AGENTS ×5 · Relationship graph derived from table definitions · Evidence search · AAF permission control
VEXPLOR AI AGENT · PURCHASE & MATERIALS
VEXPLOR AI Agent — purchasing and materials work screen
FIG.0 — AI AGENT WORKSPACEScreen data is an operating example
01WHY AGENTSERP RECORDS · HUMANS STILL DECIDE

Ten years of using the system,
and people still make the decisions.

When an order comes in, people search through ERP, PDM and Excel to check materials, set order quantities from experience, and chase suppliers by phone and email. The system only keeps the records; decision and execution are left entirely to people. These are four structural problems we have seen repeatedly on manufacturing sites.

Fragmented data

Item codes differ between ERP, PDM and outsourcing systems, so people match the same material by eye.

Many items, heavy outsourcing

The more items there are and the higher the share of outsourcing, the more complex supply decisions become, and the more they depend on one person's experience.

Delivery-date reliability

It is hard to say with confidence when materials will arrive and when production can start.

Limits of supplier management

Orders, reminders and replies for dozens of suppliers are scattered across phone calls and emails, and managing them becomes a job in itself.

02FIVE AGENTSPERCEIVE → PREDICT → AUTOMATE → COMMUNICATE → GENERATE

Five agents
work as one purchasing team.

Five agent nodes passing a packet of light on in a relayFIG — FIVE ROLES, ONE RELAY

Adding one agent per task soon turns into dozens of maintenance liabilities. How we handle that maintenance is set out in After the build. What is needed is not a number of agents but an environment in which agents can work. The point is the team structure, not the number five — five agents with roles divided from perception to generation pick up one another's results on a single ontology environment and carry the work through to the end. Purchasing and materials is the first pack — with the environment left as it is, the same approach extends to other areas such as production, quality and equipment.

C-01Purchasing analysisReads purchasing data scattered across ERP and PDM through the ontology and perceives the current status and anomalies. It finds which material is the problem before a person does.ROLE Perceive
ERP · PDM ontology mapping
C-02Supply forecastLooks at orders, stock and lead times together and forecasts when a material will run short. It gives an answer backed by evidence to delivery-date questions.ROLE Predict
Demand · stock · lead time combined
C-03Order optimisationTakes the forecast and proposes what to order, how much and when, and carries it out once approved.ROLE Automate
Proposal → approval → execution
C-04CommunicationHandles orders, reminders and replies with suppliers on your behalf, and reports only exceptions and matters for decision to people.ROLE Communicate
Automated supplier communication
C-05ReportWrites the purchasing and materials status report on its own before management asks, and reports it with evidence.ROLE Generate
Reports with evidence links attached
The time-saving scenario (BOM requirement calculation, 3 hours → 3 minutes) is a build target (plan) — not a measured result
Place in the journey · THE JOURNEY

The third step of the journey — intelligence

In the dark factory journey, this service is the third step, intelligence. Agents begin to make decisions on the data standard set up in the foundation and integration steps. The environment in which agents work starts with a diagnosis — the VEXPLOR diagnosis first checks the data standard and readiness for integration.

03ARCHITECTUREPLUG IN · DO NOT REPLACE

Existing systems stay as they are.
We connect like a plug-in.

Existing servers left as they are, with a thin agent module connecting through a single beam of lightFIG — ADD WITHOUT BREAKING

Replacing a system is a decision that takes years. AI agents do not ask for that decision — the existing ERP, MES and PDM servers are left untouched, an edge server pre-processes the data securely and passes it to the VEXPLOR AI server, and the ontology maps each system's data structure automatically.

EXISTING SYSTEMS [ ERP · MES · PDM — unchanged ] → EDGE SERVER [ pre-processing · SSL-secured transfer ] → VEXPLOR AI SERVER [ agent orchestration · evidence search · ontology ] → decision · proposal → human approval → executionFIG — SYSTEM-SAFE PLUG-IN · Automatic ontology mapping · K-SYSTEM ERP (YOUNGLIMWON) integration (semiconductor parts maker)
AS-IS — people search the systemsTO-BE — agents report to people
Sales · ordersAfter an order, material availability is checked department by department before a delivery date is givenThe supply forecast agent presents material and delivery scenarios as soon as the order arrives
BOM · material matchingItem codes in ERP, PDM and Excel are matched by eyeAutomatic ontology mapping absorbs the code mismatches; people only check the matching result
Outsourcing ordersOrder quantities are set from experience and managed by phone and emailThe order optimisation and communication agents handle proposals, dispatch and replies; people approve
04WHY TRUST ITGROUNDED · GOVERNED

An answer without evidence
is not reported.

Answers with evidence attached

An answer comes with evidence found in the company's records by following definitions and relationships. When there is no evidence it does not answer, so a plausible guess is not reported to people.

BASERelationship graph derived from table definitions · Evidence search — no answer when there is no evidence

Permission only as far as it is verified

What an agent is allowed to do is controlled through the AAF permission framework. Proposal only, or execution after approval — the company sets the scope.

CONTROLAAF L0~L4 · Approval gate · Audit log of every decision

This foundation is not only for our own agents. Whichever AI agents you come to use, working safely in a factory requires a foundation of data standards, permissions and verification to be in place first. That foundation is what VEXPLOR builds — agents will keep changing, but the foundation stays with the company.

How this technology works — technology map →
05PROOF · PARTNERSON A REAL ERP

We build
on a real ERP.

CASE

Semiconductor parts maker — manufacturing-AI specialised programme

Five purchasing and materials AI agents across multiple plants for a manufacturer of semiconductor parts (N2 Purge). The existing K-System ERP (Younglimwon) is not replaced; it is connected through an ontology.

STACKAI AGENT ×5 · K-SYSTEM ERP (YOUNGLIMWON) + PDM ontology mapping · 2026.07–2027.03
ECOSYSTEM

ERP ecosystem cooperation

A project supplying POP to the Younglimwon ERP ecosystem is delivered, and we also build under a cooperation model with ERP partners — the partner takes the customer relationship and the ERP domain, and WACE takes the AI build and operation.

MODELERP partner channel cooperation · Government manufacturing-AI specialised track can be used · Phased adoption (proposal only → execution on approval)
Programme targets (plans) are not shown
06WHEN NOT TO3 CASES

Cases where this is not a fit.

There are three. You want a fully unmanned factory right now — we commit only to step-by-step autonomy. You want office automation only, with no manufacturing — our expertise is in manufacturing. You are looking for a hardware vendor for robots or equipment — hardware is supplied by verified partners, and we take integration and operation.

Quotation basis → Equipment · integration scope · Scope of AI use · Build · integration depthFIG — An exact quotation follows the diagnosis

First, whether AI can be adopted with the systems you use today.

We diagnose the systems and data structures you use, and decide together which task's agent to start with.

Request a Diagnosis