Connecting systems produces coordinated work.
Orders, knowledge, and operating data connect the factory's context with your ERP and MES.
Start with one defined workflow. Order intake, work instructions, and shop-floor workflows run today with customer-specific integration. The overview separates available modules from the target picture.
- ContextConnect orders, knowledge, and operating data.
- Agents and toolsCoordinate tools and checks.
- Authorized actionComplete agreed work.
- FeedbackLearn from outcomes.
One module is already a product: inbox
- CNC contract manufacturer: order-entry effort fell from 20 to 2 hours weekly.
- 90 percent of entry now runs automatically.
Result from an iosys customer project, not a guarantee for other workflows.
View inbox →What to know before getting started.
What can I use today?
inbox is the packaged entry point for order intake. Order intake, work instructions, and shop-floor workflows run productively in customer projects, each with its own integration. Planning, lead time, orchestrator, knowledge, and drawing extraction are in development. An idea such as a factory-manager copilot or a predictive-maintenance agent is not a finished product unless it appears in the module list with a current status. The module list marks planned and simulated modules.
What is product and what is implementation work?
factorIQ provides reusable modules. iosys scopes the missing integration, data mapping, checks, and permissions around the outcome you choose. Cooperating agents are the target architecture, not one finished factory-wide package. We do not promise a fixed bundle or standard connectors.
How are ERP, MES and machines connected?
We begin with available APIs, database access or exports, and the permitted operations. The published cases show a cloud API for order intake, DUAP's undocumented database, and maxon's bridge between machines and ERP. Existing systems stay where feasible. Whether compatibility exists or controllers need changes is settled by an assessment. We do not promise that every older machine can be connected without changes to hardware or controls.
Where are data and AI processed?
There are three deployment options: operated by iosys in Switzerland or the EU, in your Azure, or, depending on the module, in your plant. Which one is available depends on the module. The location of the application and its data is not automatically the location where a model provider processes requests. Fully local-model production operation is not generally proven. MIND runs in production in maxon's Azure; that does not prove every module. No certification or legal assurance comes with this.
Sources:Deployment options on this pageiosys development data handling
Who approves actions and checks results?
Fixed checks, allowed read and write actions, and the responsibilities of your people are agreed per workflow. Uncertain fields and exceptions go to people. The event-log and control-point conventions are the target picture and are implemented unevenly today. There is no single audit interface across all modules, and no workflow bypasses machine safety.
What determines setup and running costs?
We do not publish a universal price for setting up or running factorIQ. These are scoped separately: interfaces and data access, workflow and checks, operating location, usage and support, and later changes. The published order-processing case at approximately CHF20,000 includes the ERP migration and is not a price list. The workshop is a separate service; its current prices are on the linked page.
How do we choose the first workflow and success measures?
Pick an available workflow with usable data and an accountable owner, not automatically a factory-manager copilot. Before a bounded pilot, agree the baseline, time per operation, review workload, error workload, and release criteria, then expand after observed performance. Data and reporting pilots can begin with read-only access; writing to the ERP needs explicit permissions. Duration is scoped case by case, not promised as a universal number of weeks. The order-processing case took three months to a result plus two months of ERP migration; DUAP documents about 30 engineering hours. Those are different measures.
What evidence supports this?
The outcomes come from iosys customer projects: the CNC order-processing case and DUAP with figures, and the productive Azure bridge at maxon. They are not all standard installations of the complete factorIQ vision. The GEWO field-confidence example is not an accuracy benchmark. Drawing-extraction results stay limited to development stage and sample size.
Sources:Order-processing caseDUAP casemaxon casePublished drawing extraction
The layers of the factory
- The layers connect context, tools, actions, and feedback.
- iosys builds missing integrations and verifiable workflows to achieve your outcome.
Tasks
Purchase orders
Work instructions
KPIs
Workstations
Review screen
Shop-floor display
Control
Fixed rules check the details. Your people decide.
Integration
ERP
MES
Documents
Operation
At your site
At iosys
In your Azure
One model for orders, parts and operations
From raw data to shared meaning
Orders, parts, operations: one modelPDF purchase order in the mailbox
Details sit as text in a PDF, not as fields in the ERP.
becomesCustomerOrderPart
Work instruction as PDF or DOC
Unstructured; every extracted field carries a confidence rating.
becomesOperationPart
Scanned drawing
No text layer, 8 of 11 sheets rotated, 7 title-block layouts.
becomesPart
ERP database
Proprietary and undocumented: about 5,000 objects narrowed to six core tables.
becomesOrderDepartment
Handwritten check marks and notes
A supplier's inspection remarks, not drawing content.
becomesSupplierPart
Promised delivery date
When a supplier moves a date, planning recalculates by hand today.
becomesSupplierPartOrder
- Customer
- Order
- Part
- Operation
- Department
- Supplier
- Customer places Order
- Order contains Part
- Part is made in Operation
- Operation runs in Department
- Supplier supplies bought-in parts Part
Today this model is built per workflow and it grows with every one. Company-wide data cleanup is not required for the first workflow.
Where does your application run?
YOUR SYSTEMS REMAIN IN CHARGE
ERP
MES
Machines
if present
factorIQ
OPERATION: IOSYS, SWITZERLAND OR EU
OPERATION: IN THE CUSTOMER'S AZURE
OPERATION: ON YOUR PREMISES
Order entry
PDF to ERP
runningPlanning
Material, due dates
in developmentWork instructions
PDF to MES
runningShop floor
one screen per department
runningCondition
Motor data, good / bad
runningCause analysis
explains deviations
plannedUtility duct
Event log · Model gateway · Control points · MCP to the outside
WHERE PEOPLE WORK
Own interface
Checks, dashboards
Teams, Copilot
connected via MCP
Back into the ERP
and the MES
Cloud model
hosted by iosys, in Switzerland or the EU
Azure OpenAI
in your tenant
Local model
Target, not yet proven in production
Documents and order data reach iosys servers in Switzerland or the EU.Everything runs in your own Microsoft environment.A local AI model is not yet proven in production.
The order flow in detail
Model choice
Cloud, tenant or local
ConfigurationSuppliers
Bought-in parts, dates
Customer
orders and receives
Order intake
read, check, create
runningLead time
Bought-in parts in the BOM
in developmentPlanning
Capacity and due dates
in developmentOrchestrator
classifies, forwards
in developmentERP and MES
Orders, BOMs
Customer systemShipping
Goods and documents
simulatedManufacturing
Machines and feedback
simulatedShop floor
KPIs per department
runningMCP integration
Teams, Copilot, Work IQ
InterfaceOrder flow
Customer
orders and receives
Order intake
read, check, create
runningPlanning
Capacity and due dates
in developmentManufacturing
Machines and feedback
simulatedShipping
Goods and documents
simulatedSupporting work
Suppliers
Bought-in parts, dates
Lead time
Bought-in parts in the BOM
in developmentOrchestrator
classifies, forwards
in developmentShop floor
KPIs per department
runningSystem integration
Model choice
Cloud, tenant or local
ConfigurationERP and MES
Orders, BOMs
Customer systemMCP integration
Teams, Copilot, Work IQ
Interface- Legend
- factorIQ module
- Interface
- Customer system
- Model
- Integration
- External
- Order flow: solid; supporting work: thin; system integration: dashed.
- Colors show the type of module.
Modules and their current status
Available and in use
Order entry
available as SaaSThe model reads PDF purchase orders from the mailbox; deterministic checks validate the result, and a person approves. The approved result enters the ERP in structured form.
Evidence: 20 → 2 hours per week at a CNC contract manufacturer
to the product →Work instructions
running at customersUnstructured instructions from PDF and DOC are extracted and entered into the MES, with a confidence rating per field and a check by the operator.
Evidence: in production, export to the Hummingbird MES
Shop floor
running at customersKPIs from ERP and MES where the work happens: one screen per department, orders, due dates, and deviations as soon as they occur.
Evidence: ten screens in around 30 hours of engineering effort
In development
Planning
in developmentStock, bills of materials, and capacity over six months. Changes remain a preview until the planner confirms them.
Evidence: demonstrable version with preview, confirmation, and reset
Lead time
in developmentTracks the bought-in parts from the bill of materials: promised dates, delays, stock. Reports bottlenecks before they show up in manufacturing.
Orchestrator
in developmentThe module receives, classifies, and forwards. Fixed rules and people decide.
Knowledge
in developmentManuals, work instructions, and file stores can be queried. Every answer names its source.
Planned
Root-cause analysis
plannedExplains deviations from measurements, machine data, and tool changes. A person chooses the action.
Service
plannedQueries after delivery: order, drawing, and inspection report pulled together, with a draft reply ready for approval.
What connects the modules
Four conventions for the design
- Current installations implement the four conventions unevenly.
- A product-wide event interface is missing; details identify each deployment's status.
Event log
- Modules record actions with a reason and a source.
- Interfaces read only events; module internals remain separate.
Model choice
- Modules use shared model selection instead of direct calls.
- Cloud, customer-tenant, and local computation are configured.
Control points
- Each module has at least one human decision point.
- Models interpret uncertainty; deterministic tools check calculated results.
External integration
- MCP exposes capabilities for your AI environment.
- Teams, Copilot, and Work IQ can use these tools.
Three deployment options, the same software
- Which deployment option is available depends on the module.
Operated by iosys
- iosys runs the application in Switzerland or the EU.
- You use your own review screen.
In your Azure
- The application runs in your Microsoft environment. maxon runs MIND there in production.
- This does not prove every module. Teams/Copilot integration is partly proven.
In your plant
- Production data stays in the plant.
- Complete production operation with a local AI model remains unproven.