AI-native · MES development · Production

The AI-native MES
development platform.

OneHub MES builds one kind of system: MES — Manufacturing Execution System software. Ask for a change inside the MES your operators already use; a working release comes back the same shift.

How AI-native development works

Six stages. One shift, from a sentence to a release.

Stage 00 is the MES already running in production. Stages 01–06 turn one request into a release — in a single shift.

mes.nordwind.example/production/wo/WO-2027-0114-07
STAGE 00 · BASELINE
Home
Master data▾
Production▴
Order execution
Line monitor
Quality▾
Batch trace▾
OEE & SPC▾
Notifications
Production / Order execution · Ingolstadt · Line 2
PRODv2.8.1R. Hoffmann · Operator
Order executionWO-2027-0114-07 · shift A · 3 functions only
Print labelReport quantity
WO-2027-0114-07
EP-Resin A · 480.000 kg
Line 2 · Compounding
REC-EP-A · rev 07
120.000 kg
R. Hoffmann
3
Operations
4
Components in recipe
120.0kg
Reported so far
0
Weight verifications
OpDescriptionStationPlannedReportedStatus
0010Weigh-outScale S2480.000 kg120.000 kgIn progress
0020MixMixer M240 min—Waiting
0030Discharge & packPack line P115 min—Waiting
Showing 3 of 31
Three functions on this screen: start an operation, type the reported quantity, print the label. The material is whatever the operator picked from the dropdown and the scale reading is keyed by hand — nothing compares either of them to the recipe.
Operation record✕
Operation
—
Description
—
Station
—
Planned
—
Reported
—
Status
—
Operator R. Hoffmann · started 10:22
Labels printed 3 labels · no scan on this operation
Actions available: 3 (Start · Report · Print) · no weighing screen exists
Home
Master data▾
Production▴
Order execution
Line monitor
Quality▾
Batch trace▾
OEE & SPC▾
Notifications
Production / Order execution · Ingolstadt · Line 2
PRODv2.8.1R. Hoffmann · Operator
Order executionWO-2027-0114-07
Print labelReport quantity
EP-Resin A · 480.000 kg
Line 2 · Compounding
REC-EP-A · rev 07
120.000 kg
3
Operations
4
Components
120.0kg
Reported
0
Verifications
OpDescriptionStationPlannedStatus
0010Weigh-outScale S2480.000 kgIn progress
0020MixMixer M240 minWaiting
0030DischargePack line P115 minWaiting
Wrong-material batches, last 90 daysall found downstream
BatchComponentStationWhat went inFound at
B-2611 0410 · EP-Resin AS2EP-Resin BQA gate
B-2612 0930 · Filler CS2118.700 kgEnd of mix
B-2701 0320 · Curing agent BS3lot expired 12.12Pack line
Showing 3 of 31
The operator is standing on this screen. The change has to land here — not in a chat window somewhere else.
Operation record✕
Operation
—
Description
—
Station
—
Planned
—
Reported
—
Status
—
OneHubask, or drop a file
Context attached · Production / Order execution · WO-2027-0114-07 · v2.8.1 · role OPERATOR
PDFSOP_v4.pdf✓ XLSXdev_log.xlsx✓ PNGerror.png✓
R. Hoffmann10:31
We ran three wrong-material batches on this line last month. When a material is scanned, check it against the recipe component, and check the scale reading — wrong material, or more than ±1 % off, lock the station and make a shift lead release it with a reason.
OneHub10:32 · structured
Four rules, six fields, one new screen, bound to ISA-95 ProcessSegment and IATF 16949 §8.5.1.
R1Scan resolves a MaterialLot, must equal the recipe component
R2Scale reading vs theoretical qty, deviation % stored per component
R3Mismatch or |dev| > 1 % → station LOCKED, Report blocked
R4SUPERVISOR releases with a reason; both actions audited
From your files: SOP §4.2 already sets ±1 % (p. 14); the screenshot is the existing error dialog, so I reuse it.
Net change: +1 menu · +1 page · +6 fields · +2 buttons · +1 rule · +2 records
Try one Reject an expired lot at scan
⬆Drop a file or a photo of the screen — PDF, XLSX, PNG · 25 MB
Describe the change…
+ AttachSend
Change scope — scan-verified weigh-outreuse 71%3 functions in, 5 out, 1 new screen
ObjectNow · v2.8.1After this run · v2.9.0Delta
Menu entriesOrder execution · Line monitor+ Weigh-out (weighing verification)+1 menu
Screens1 — order execution+ Weighing verification (own page, reachable only from the new menu)+1 page
Fields on the operationmaterial · reported qty+ theoretical qty · actual qty · deviation % · lot · scan time · release reason+6 fields
ButtonsStart · Report · Print+ Scan material · + Supervisor release+2 buttons
Rules—recipe match + ±1 % guard, station lock, release with reason+1 rule
Recordsbatch record+ scan event log · + deviation / release record+2 records
Unchanged order release, ERP interface, label printing, batch-record layout, the operator's menu. Out of scope automatic dispensing hardware and in-line NIR identification (decision D-03, deferred).
Process blueprint — weigh-out with verification6 nodes · 2 branchesreuses the existing operation lifecycle
N1
Operation start
N2 · new
Scan material
N3 · new
Recipe match
N4 · new
±1 % check
N5
Report posted
Branch A material matches and deviation ≤ 1 % → component released, next component  ·  Branch B mismatch or deviation > 1 % → station LOCKED, Report blocked, shift lead notified
Trigger scan event on station S2 · Idempotency key order + operation + component + lot · Clock scale settles for 800 ms before the reading is taken (D-02)
NodeOwnerSystem actionSLA
N2 Scan materialstation serviceresolve MaterialLot, bind to operation≤ 300 ms
N3 Recipe matchrule enginecompare lot.materialNo to ProcessSegment component≤ 50 ms
N4 ±1 % checkrule enginedeviationPct = (actual − theoretical) / theoretical, lock if |dev| > 1≤ 50 ms
B Lock & releaseshift leadreason code + free text, audit entry, unlockhuman
ISA-95 mapping — the same objects the plant already runsno new master dataIEC 62264-2
ObjectISA-95 entityTableChange
Work orderProductionOrdermes_prod_orderread-only
OperationProductionOrderSegmentmes_op+ 3 columns
Recipe lineProcessSegment / MaterialSpecmes_recipe_compread-only
Material lotMaterialLotinv_lotread-only
VerificationMaterialLot + Personnelmes_weigh_verifnew table
ReleasePersonnel / auditaud_event+ 2 event types
Why it matters the new screen reads the same BOM, lot and person records as the rest of the plant — no shadow master data, no reconciliation job, no second place to look when the numbers disagree.
Field spec — six new fields, three on the existing operationgenerated from R1–R4checked at the API boundary
FieldTypeRule
theoretical_qtydecimal(10,3)copied from the recipe component at operation start, never editable
actual_qtydecimal(10,3)taken from the scale when the reading settles; 0 < actual ≤ 1.2 × theoretical
deviation_pctdecimal(6,3)generated column; lock when |dev| > 1.000
lot_iduuid → inv_lotmust resolve; expired lot rejected with LOT_EXPIRED
lock_stateenumOPEN · LOCKED · RELEASED — transitions audited
release_reasontextmandatory when a shift lead overrides; ≥ 12 characters
Permissions — who may do whatexisting IdP groupsno new role created
RoleScanOverride / releaseChange the ±1 % limit
OPERATOR✓——
SHIFT_LEAD✓✓ with reason—
PROCESS_ENG✓✓ with reason✓ (change request)
SYSTEMwrites the locknevernever
Open decision D-02 the scale settle window is 800 ms; line 2 has an older indicator that may need 1 200 ms. Confirm with maintenance before release.
Run 0114-03 · 10:30 → 11:19 4 agents in parallel PRJ-NWC-MES-24
49minutes
wall clock, inside one shift
1h 44magent time
across 4 parallel lanes
2human gates
6 min of people time
12files
2 940 lines · 3 new tables
TaskTypeAgent timeLane10:30 → 11:20 · 5-min columns
Weighing entity + order read-modelbackend25 mA
Recipe component resolver (BOM → segment)backend18 mA
Scan endpoint + MaterialLot lookupbackend22 mB
Deviation guard + station lock rulebackend14 mB
Weighing verification page (new)UI12 mC
Order execution: +6 fields, +2 buttonsUI8 mC
OQ scripts + trace matrixverify3 mD
Runbook + changelog for your ITdocs2 mD
Shift-lead walkthrough on line 2gate—human
build agent · running
src/mes/production/
weigh-out.rule.ts +96
weighing.service.ts +168
weighing.controller.ts +72
recipe.resolver.ts +54
deviation.guard.ts +41
db/migrations/
0058_weighing.up.sql +34
ui/pages/
weighing.page.vue +212
order-execution.page.vue +48
policy/
weighing.rego +26
… writing file 1 of 6
PR #418 feat(mes): scan-verified weigh-out with a ±1 % station lock backend agent
1// from FDS-0114-03 · rule WGH-SCAN-01 · reviewed by A. Klein
2import { Rule, RuleResult } from '@onehub/rule-engine';
3
4@Rule({ code: 'WGH-SCAN-01', clause: 'ISA-95.ProcessSegment', tolerancePct: 1.0 })
5export class WeighOutVerification implements Rule<MesCtx> {
6 async evaluate(ctx: MesCtx): Promise<RuleResult> {
7 const { op, lot, component, reading } = ctx;
· lint· typecheck· unit 0/142· migration dry-run· SBOM2 approvals pending
Operational qualification · run #41 56 / 56 passed traceability 100% Verify agent · 1 m 38 s
click a test for its steps
✓OQ-06Correct material scanned → verified, Report unlocked0.6s
✓OQ-09Wrong material scanned → station LOCKED, Report blocked0.7s
✓OQ-14Scale reading +1.35 % → LOCKED, deviation stored0.5s
✓OQ-17Shift-lead release with reason → unlocked, audited0.9s
✗OQ-22Boundary: exactly +1.000 % must still pass0.8s
✓OQ-25OPERATOR may not release a locked station (403)0.3s
✓OQ-29Expired lot at scan → LOT_EXPIRED, no record written0.4s
✓OQ-33Rescan on a locked station is idempotent0.5s
✓OQ-38QA hold on a lot blocks the scan with LOT_ON_HOLD0.4s
✓OQ-56End to end: release → scan ×4 → lock → release → report3.1s
OQ-06 · correct material verified
Linked URS
URS-01 · IATF 16949 §8.5.1
Steps
1. Start op 0010 on WO-2027-0114-07
2. Scan lot L-4419-02 for component 10
3. Read the verification row
Expected
lock_state OPEN, deviation stored, Report enabled
Actual
Matched · 0.6 s
OQ-09 · wrong material locked
Linked URS
URS-02 · IATF 16949 §8.5.1 (error proofing)
Steps
1. Scan a lot of EP-Resin B on component 10
2. Attempt Report quantity
Expected
409 WRONG_MATERIAL, station LOCKED, Report returns 423
Actual
Matched · 0.7 s
OQ-14 · deviation above the limit
Linked URS
URS-03 · recipe tolerance ±1 %
Steps
1. Theoretical 120.000 kg
2. Reading 121.620 kg (+1.35 %)
3. Read the verification row
Expected
lock_state LOCKED, deviation_pct 1.350
Actual
Matched · 0.5 s
OQ-17 · release with a reason
Linked URS
URS-04 · 21 CFR 11.10(d) · Annex 11 §9
Steps
1. Lock the station
2. Release as SHIFT_LEAD with a 24-character reason
3. Read audit events
Expected
lock_state RELEASED + STATION_RELEASED entry carrying the reason
Actual
Matched · 0.9 s
OQ-22 · boundary at exactly 1 %
Linked URS
URS-03 · decision D-02 (tolerance inclusive)
Steps
1. Theoretical 100.000 kg
2. Reading 101.000 kg
3. Read lock state
Expected
OPEN — the limit is inclusive, the operator is inside tolerance
Actual
Locked at exactly 1.000 % — comparison used rounding to 2 dp, 1.004 > 1
Defect
DEF-047 fixed: compare on decimal(6,3) before rounding; retest passed, same run
OQ-25 · operator cannot release
Linked URS
URS-04 · segregation of duties
Steps
1. Call release as OPERATOR
2. Call release as SHIFT_LEAD
Expected
403 for OPERATOR, 200 for SHIFT_LEAD
Actual
Matched · 0.3 s
OQ-29 · expired lot rejected
Linked URS
URS-02 (material identity)
Steps
1. Scan a lot with expiry 2026-12-31
2. Count verification rows
Expected
409 LOT_EXPIRED, zero rows written
Actual
Matched · 0.4 s
OQ-33 · rescan idempotency
Linked URS
URS-01 (data integrity)
Steps
1. Scan the same lot twice while locked
2. Count rows
Expected
One row, second call returns the first
Actual
Matched · 0.5 s
OQ-38 · QA hold blocks the scan
Linked URS
URS-02 · ISO 9001 §8.7
Steps
1. Put lot L-4301-08 on QA hold
2. Scan it on component 30
Expected
409 LOT_ON_HOLD, message shown on the station screen
Actual
Matched · 0.4 s
OQ-56 · end to end
Linked URS
URS-01…10 (whole chain)
Steps
1. Release the order
2. Scan and weigh four components
3. Force one deviation
4. Release as shift lead
5. Post the report
Expected
Complete batch record with 6 audit entries and a signed pack
Actual
Matched · 3.1 s
56OQ scripts
1defect found · 1 fixed
requirement coverage
10/10URS
Home
Master data▾
Production▴
Order execution
Weigh-outNEW
Line monitor
Quality▾
Batch trace▾
OEE & SPC▾
Notifications
Production / Weigh-out · op 0010 · Ingolstadt
PRODv2.9.0R. Hoffmann · Operator
AI iteration 1New menu + screen: Weigh-out 2+6 fields and Scan on the operation 3Station lock with supervisor release v2.8.1 → v2.9.0 · 3 changes, traced to URS-01…10
Weighing verification · op 0010WO-2027-0114-07 · REC-EP-A rev 07 · scale S2
+ Scan materialReport quantity
480.000kg
Theoretical, 4 components
478.930kg
Actual, weighed
−0.22%
Order deviation
1
Component locked
CompMaterial · lotTheoreticalActualDev %Check
10EP-Resin A · L-4419-02240.000240.180+0.08Verified
20Curing agent B · L-4392-1196.00096.050+0.05Verified
30Filler C · L-4301-08120.000118.700−1.08Locked
40Additive D · —24.000——Pending
Station S2 locked. Component 30 Filler C is 1.313 kg below theoretical (−1.083 %). Report quantity is blocked until a shift lead releases the station with a reason. Supervisor release is now a button on this screen — the release, its reason and the operator are all written to the batch record.
4 components · scan event log 3 · deviation record 1 · batch record updated2.9.0
Verification record✕
Component
—
Material · lot
—
Theoretical
—
Actual (scale)
—
Deviation
—
Check
—
Scan 10:22:41 by R. Hoffmann
Fields added by this iteration: theoretical · actual · deviation · lot · scan time · release reason
MES only — no general apps6 design partners, Q1 2027No credit card
87
production plants running our systems
9
manufacturing verticals
30+
patents and IP assets
1:1
requirement-to-test traceability
Positioning

Where general-purpose tools stop.

MES is the only thing OneHub builds. Coding assistants make one developer faster; app builders produce a prototype in an afternoon. Neither knows what a material lot, a recipe revision or a shift handover is.

Runs in productionEvery run ends in a merged, tagged release in your repository — not a link to a sandbox that dies when the tab closes.
Yours to changeSource, migrations and tests land in your repo. Your own process engineers change the system without raising a purchase order.
Built for a plantMulti-site, multi-org, behind your IdP, deployed by your CI — sized for an engineering organisation, not a demo day.
AI coding assistants
Cursor · Copilot · Windsurf
AI app builders
Lovable · Bolt · v0
Custom development / SI
Project delivery
OneHub
Built forIndividual developersNon-technical foundersOne-off programmesManufacturing IT, OT & quality
Domain modelNoneNoneReinvented per projectISA-95 / ISA-88 native
Management logicNoneNoneWritten from scratchModelled per module
OutputCode suggestionsA working demoSystem + black boxSystem, source, migrations, tests
Who changes it nextThe developerThe platform vendorWhoever gets the change orderYour process engineers
How a change shipsRe-prompt until it looks rightRebuild the projectA new statement of workPR, tests, tag, rollback
Runs onDeveloper laptopVendor platformVendor or clientYour cloud, your DC, your keys
Deployed byThe developerThe platformThe vendorYour IT team
Multi-site, multi-org——Depends on budgetBuilt in
Time to first release—Hours (demo)6–18 monthsOne shift

General-purpose tools stop at “it runs.”
OneHub starts there — and keeps going.

Why it lands differently

Flexible, fast, yours to change.
And built to survive Monday morning.

A plant does not need another MES it cannot adjust. It needs one that bends with the process, ships before the shift ends, and holds up at three in the morning.

01 / FLEXIBLE

Change the process, not the project plan.

Routes, batch rules, inspection points, approval levels — they live in the model, not in a backlog. A new check at one station does not wait for next quarter.

02 / FAST

A change lands the same shift.

One request, one release: minutes of agent work, a short review, a merge. What used to be a change request with a quote becomes an afternoon.

03 / YOURS TO RUN

Your engineers do it themselves.

The process engineer who owns the line, and the IT team who owns the deployment — not a vendor queue. No change order for a single field.

04 / PRODUCTION-GRADE

Not a demo that impresses once.

Transactions, migrations, automated tests, a rollback that has been run. Nothing reaches the line that has not run in staging first.

Under the hood

Why it behaves like an MES,
not like a code generator.

An AI is only useful here if it knows the domain, understands how the work is managed, and ships like engineering software.

It knows the plant

Batch chemical, pharma, food, electronics, automotive Tier-1 — the schema and the exception paths are the product. It already knows what a material lot or a line clearance is, so you describe the change, not the data model.

It models the management, not the screens

An MES is a management system before it is an interface: who releases an order, what a batch records, when a deviation escalates. Modelled on ISA-95 and ISA-88, carried per module.

It ships like engineering

Every change is a pull request: reviewed, tested, versioned, reversible. Read the diff before it touches the line; roll it back in one command.

The management model

We did not invent an MES data model.
We adopted the one plants already use.

ISA-95 and ISA-88 are not certificates to hang on a wall. They are the vocabulary a planner, a shift lead and a plant manager already share — and OneHub puts it in the schema.

ISA-95 / IEC 62264object model

Equipment, material lot, personnel, schedule and production performance as first-class entities — not tables someone invented on site.

MODELLED · SCHEMA
ISA-88 / IEC 61512batch control

Recipes, unit procedures, operations and phases, with the state model a batch engineer expects.

MODELLED · SCHEMA
B2MMLERP ↔ MES

Message contracts for production orders, material lots and performance, so integration is a mapping exercise, not an invention.

MODELLED · CONTRACTS
OPC UAshop floor

Tag ingestion with context: which line, which batch, which shift. Data without context is a spreadsheet.

MODELLED · CONNECTORS
ISO 9001:2015§8.5 · §10.2

Nonconformity, corrective action and process control flows generated from the clause requirements you actually claim.

MODELLED · WORKFLOWS
IATF 16949automotive

APQP/PPAP artifact handling, control plans and lot-level genealogy when a Tier-1 customer asks where a part came from.

MODELLED · TEMPLATES
21 CFR Part 11e-records

Electronic signatures, immutable audit trails, record retention and access control — generated at the data layer, not bolted on.

MODELLED · DATA LAYER
EU GMP Annex 11computerised systems

Change control, periodic review, incident logging and data integrity by design.

MODELLED · CONTROLS
GAMP 5 (2nd ed.)risk-based CSV

URS → risk assessment → test plan, produced as the pipeline runs rather than reconstructed afterwards.

MODELLED · LIFECYCLE
ISO 13485medical devices

Device history records, batch release gates and design-transfer traceability for regulated production.

TEMPLATE
ISO/IEC 17025laboratory

LIMS result integrity, method traceability, analyst competence and instrument calibration records.

TEMPLATE
ⓘ
Standards are a shared language, not a promise. OneHub models the management logic those standards describe, and gives your engineers a system that matches the way they already work. Interpreting them for your plant — and signing for it — stays with your own people.
Modules

One object model, six modules —
each built on its own standards.

MES is not one thing. Production answers to ISA-95, quality to ISO 9001 and IATF 16949, the laboratory to ISO/IEC 17025, records to GMP Annex 11 and 21 CFR Part 11. OneHub keeps a single ISA-95 object model underneath and attaches the right management logic to each module — so a nonconformance and a lab result are handled the same way, without pretending they answer to the same rule.

Module What it covers Standards the module is built on
MES
production
Production orders, material flow, lot genealogy, OEE, shift handover, operator guidance ISA-95 / IEC 62264 · ISA-88 / IEC 61512 · B2MML · OPC UA
QMS
quality
Nonconformity, CAPA, deviation, change control, supplier quality, batch release ISO 9001 §8.5 / §10.2 · IATF 16949 · 21 CFR Part 11
LIMS
laboratory
Sampling plans, test methods, result approval, instrument calibration, analyst competence ISO/IEC 17025 · GMP Annex 11 · ALCOA+ data integrity
WMS
warehouse
Goods receipt, put-away, line-side supply, tank farm, expiry and re-test control ISA-95 material model · GMP Annex 11 · EU GMP Annex 1
EAM
equipment
Equipment master, maintenance plans, calibration and PM, downtime and reliability ISA-95 equipment model · ISO 55001 alignment
SPC
analytics
Control charts, process capability, batch trending, release by exception ISO 7870 · ISO 22514 · GAMP 5 (risk-based)

Binding depth is agreed in the URS for each project — we would rather model four clauses properly than claim thirty. Anything not supported today is listed as a template, not sold as a feature.

Platform

Three layers, one artifact chain.

Other platforms give you a model and a chat box. OneHub gives you the MES model, the engineering pipeline that produces against it, and the agents that do the work — with a human approval gate at every irreversible step.

LAYER 01
Standards
What a manufacturing system must be
ISA-95 model
Objects & relations
Clause library
Standards → controls
Industry templates
9 verticals
UI system
Dense, operable, i18n
LAYER 02
Engineering
How the system gets built and proven
6-stage pipeline
Intake → release
Traceability
requirement ↔ model ↔ test
Test pack
OQ · trace matrix
Release governance
PR, tag, rollback
LAYER 03
Agents
Who does the work, and who approves it
Intake
Slack · Teams · email
Blueprint
Spec & mapping
Build
Code, tests, IaC
Verify
OQ & evidence
LAYER 00
Your environment
Your account, your keys, your call
Your account
AWS · Azure · GCP · private DC
Your identity
SAML · OIDC · SCIM
Your repo
GitHub · GitLab · Azure DevOps
Your pipeline
Actions · Jenkins · Argo
Deployment

Public cloud, private cloud,
or your own data centre.

Deploy it where the data already lives: your cloud account, your bill, your model token. OneHub never holds credentials to the environment it builds into.

01
Your account, your tenancy

Runs on your cloud subscription or your own hardware. No shared tenancy, no data crossing a boundary you did not approve.

02
Your model account and token

Connect the model access you already have, with your own keys, or point the agents at a model you host. The cost, the rate limits and the data terms stay yours to control.

03
Where the data sits

EU, US or APAC pinning, a private cloud inside your network, or fully on-premises and air-gapped for plants that never see the public internet.

04
Your team runs it

We generate the release; your pipeline ships it. Read every diff before it lands, and roll back with one command if it does not behave.

AWS
Your AWS accountTerraform · EKS · RDS · EU / US regions
AZ
Your Azure tenantBicep · AKS · EU West · private endpoints
GCP
Your Google Cloud projectGKE · Cloud SQL · regional pinning
PC
Private cloud or on-premisesVMware · OpenShift · Helm · air-gapped
SSO
Your identity providerEntra ID · Okta · Ping · Keycloak
CI
Your delivery pipelineGitHub Actions · GitLab · Azure DevOps
Fit

Deliberately narrow.

We build MES — nothing else. Better to be clearly wrong for some teams than vaguely right for all of them.

Built for
  • Manufacturing IT / OT leads replacing a 15-year-old MES or consolidating four plants onto one model
  • Quality teams whose process has outgrown the spreadsheet the workaround lives in
  • Operational excellence groups rolling out standard work across sites
  • System integrators and MES consultancies who want to deliver in weeks without diluting quality
Not built for
  • Marketing sites, portals or internal admin tools — use a general-purpose builder, it is cheaper
  • Teams without a plant, a process owner or a quality system
  • Anyone looking for a no-code tool that operators configure on their own
  • Projects where nobody will own the deployment

Get early access.

We are onboarding a small group of design partners before general availability. If you are replacing an MES, standardising across plants, or delivering MES projects as a partner — we want to talk.

  • 1
    A 30-minute call to understand your plant, your standards and your constraints.
  • 2
    A working run on one of your real requirements — not a canned demo.
  • 3
    Design partner terms: no licence fee during the preview, direct line to the product team.
Prefer email? Write to chenkai@xsmartdata.com — a person answers, usually within two business days.
⚠ No submission endpoint configured — this form is running in local demo mode.
You're on the list.

Thanks — we'll come back to you within two business days from a real person, with two or three questions about your plant and your standards.

Reference: —

Questions

The ones we actually get asked.

What does "AI-native MES development platform" mean?
It means the MES is engineered by agents against a manufacturing model, not typed by hand against a blank editor. You ask for a change inside the screens your operators already use; agents draft the specification, the code and the tests; the model they work against is ISA-95, the management logic is ISO 9001 / IATF 16949 / GxP, and a person signs at every gate. It is not a coding assistant (no domain model), not an app builder (no deployment discipline) and not low-code (the output is ordinary code in your repository).
Is OneHub only for MES?
Yes. OneHub is scoped to manufacturing execution and the systems that sit next to it — MES, QMS, LIMS, WMS, EAM, SPC. It is not a general-purpose app builder, and it will decline work outside that scope: ask it for a marketing site or an HR portal and the honest answer is that a general-purpose tool will serve you better and cheaper. That narrowness is the product, not a limitation — the object model, the vocabulary and the exception paths only work because they are not shared with anything else.
Is this a low-code tool?
No. OneHub generates ordinary, reviewable code — TypeScript services, SQL migrations, RBAC policies, Helm charts — into your repository. There is no proprietary runtime and no visual editor your operators are expected to master. The difference is that the generation is constrained by a manufacturing model and a standards library, so the output is a system rather than a draft.
How is this different from Cursor, Copilot or Lovable?
Those tools optimise for the moment code appears. They have no idea what a material lot is, what clause 8.5.1 requires, or that a weigh-out has to be checked against the recipe component before the report is posted. OneHub is vertical: the domain model, the management logic and the release discipline are the product. If you are building a landing page, they are better and cheaper. If you are building an MES, they stop exactly where your project begins.
Does this replace my system integrator?
No, and we would be suspicious of anyone claiming to. Integrators own process design, change management and the relationship with the plant. OneHub removes the part of their work that is pure rework: rebuilding the same object model, re-deriving the same requirements, re-assembling the same test evidence. Several of our design partners are integrators.
Where does the software actually run?
In your environment. Your AWS account, your Azure tenant, your GCP project, or on-premises and air-gapped. Your identity provider handles authentication, your repository holds the code, and your CI/CD pipeline performs the deployment. OneHub does not hold production credentials.
Which standards are supported today?
ISA-95 and ISA-88 are native to the data model, and ISO 9001, IATF 16949, 21 CFR Part 11, EU GMP Annex 11 and GAMP 5 are bound at clause level in the current build. ISO 13485, ISO/IEC 17025 and EU GMP Annex 1 exist as templates in design-partner scope. Anything not on the list we will tell you honestly rather than model badly.
How long until we can run a real requirement?
A design partner's first run — one requirement, blueprint to a merged release in their own environment — typically lands in the first two weeks, because the model and the pipeline already exist. Broader rollout depends on your data quality and how quickly your own team can absorb the change, not on our generation speed.
What happens to our data?
Your production data never reaches us. Requirements and specifications used during a run are processed under the design-partner agreement, stored in the region you choose, and are never used to train shared models. Export and deletion are available on request.
What does it cost?
Design partners pay no licence fee during the preview. At general availability, pricing is per site with an enterprise tier for multi-plant standardisation. We will publish it rather than hide it behind a form — the number is not the interesting part of the conversation.