Data

Tools

Indexes

Applications

Experiments

Awesome

An open API service indexing awesome lists of open source software.

https://github.com/baryonlabs/open-mes-korea

Multilingual open-source MES for shop-floor execution, LOT traceability, manufacturing data collection, modular AI, and on-premises deployment.
https://github.com/baryonlabs/open-mes-korea

ai-governance digital-twin industrial-iot industry-4-0 korea lot-traceability manufacturing manufacturing-data manufacturing-execution-system mes oee open-source predictive-maintenance shop-floor smart-factory

Last synced: 16 days ago
JSON representation

Multilingual open-source MES for shop-floor execution, LOT traceability, manufacturing data collection, modular AI, and on-premises deployment.

Awesome Lists containing this project

README 

          



# Open MES Korea



**A small, robust, and extensible open-source MES**



Built from real Korean manufacturing workflows and designed to extend across

countries, languages, and industries.



[한국어](README.md) · [日本語](README.ja.md) · [简体中文](README.zh-CN.md)



[Vision](docs/vision.md) · [MVP](docs/mvp-scope.md) ·

[Architecture](docs/system-architecture.md) · [Roadmap](docs/roadmap.md) ·

[Contributing](CONTRIBUTING.md)






> [!IMPORTANT]

> Open MES Korea is currently in the design and early implementation stage.

> It is not yet a production-ready product, and its APIs and documentation may

> change during development.



![Open MES Korea - from manufacturing data collection to analysis](site/assets/open-mes-factory-team.jpg)



## Why Open MES Korea?



Manufacturing data is often scattered across work orders, paper reports,

spreadsheets, barcode systems, equipment, and ERP software. Even factories

with existing systems can struggle to connect process progress, defects, and

material-to-product lot history.



Open MES Korea does not aim to rebuild an entire ERP. It focuses on:



```text

Shop-floor execution

  + Lot traceability

  + Manufacturing data collection and analysis

  + Governed AI

```



Korean manufacturing terminology, practices, and deployment constraints are

the practical starting point. The architecture, however, is designed for

multilingual interfaces, regional terminology, and industry-specific

extensions.



## Preserve the Data Golden Time



The first step toward manufacturing AI is not selecting a model or analytics

vendor. It is collecting today's shop-floor data before it is lost.



Open MES Korea starts with operator input, barcodes, CSV, and standard

ingestion APIs. Data remains portable through PostgreSQL, CSV, and APIs, so a

factory can later compare AI modules and vendors using its own history instead

of restarting data collection after a long selection process.



## Design Goals



- **Small core:** Keep only common manufacturing execution workflows built in

- **Multilingual:** Korean by default, with support for more languages

- **Extensible:** Connect ERP, equipment, analytics, and industry rules

- **Reliable:** Audit logs, idempotency, correction history, and recovery

- **Robust:** Preserve production data integrity and explicit lot genealogy

- **Fast:** Keep shop-floor input and operational queries responsive

- **Operator-first:** Prioritize tablets, barcodes, and short input flows

- **Safe AI:** Separate reading, suggestions, approval, and execution



## Deployable Data Sovereignty



Open MES Korea is designed as installable open-source software, not a SaaS

service that requires production data to be sent to a specific cloud.



- Run the MES and database inside the factory network or a private cloud

- Keep production, lot, quality, and equipment data in operator-selected storage

- Support restricted internet access and segmented OT networks

- Enable external AI, email, analytics, and ERP integrations only by configuration

- Keep PLC command write disabled by default



A default installation does not intentionally send production data to an

external service or control equipment. Open source and on-premises deployment

do not guarantee security by themselves; firewalls, identities, TLS, secrets,

backups, patching, and installed extensions must still be operated correctly.



## Core Workflow



```text

Items / BOM / Processes

→ Work order

→ Operation start and completion

→ Production and defect records

→ Material lot consumption

→ Product lot creation

→ Production status and lot traceability

→ Basic analytics and AI queries

```



## Core and Extensions



### Core



- Master data, work orders, operations, production results, and defects

- Material consumption, product lots, and lot genealogy

- Manual input, barcode, CSV, and authenticated HTTP ingestion

- Production quantity, defect rate, and operation-time analytics

- Users, permissions, audit logs, and event outbox

- REST APIs, webhooks, and standard manufacturing events



### Extensions



- ERP, WMS, QMS, and external-system integrations

- MQTT, OPC UA, Modbus, serial, and PLC connectors

- High-throughput ingestion with Phoenix Broadway

- Telemetry analysis with TimescaleDB or ClickHouse

- OEE, advanced planning, predictive analytics, and industry dashboards

- Barcode and label templates and site-specific workflows

- AI queries, recommendations, and approval-based actions



Site-specific requirements should use explicit extension points instead of

continuously expanding the core.



## Manufacturing AI Modules



The extension roadmap goes beyond predictive maintenance:



- Vision inspection for surface, dimensional, assembly, and internal defects

- Virtual metrology and AI-assisted digital-twin commissioning

- Process-variable, yield, and energy optimization

- AMR/AGV routing and vision-guided robotic picking

- Demand, inventory, and APS scheduling optimization

- Predictive maintenance and equipment anomaly detection

- PPE, restricted-area, and hazardous-behavior detection

- Manufacturing LLMs for manuals, work instructions, and troubleshooting



These capabilities are optional modules, not built-in core features. Control,

quality disposition, and worker-safety decisions require explicit policies,

fail-safe behavior, site validation, human approval, and audit records.



## Data Collection and Analysis



The project aims to let factories collect and analyze their own operational

data without requiring a separate analytics platform first.



The default scope includes:



- Operator input, barcodes, CSV, and authenticated HTTP APIs

- Validation for malformed, duplicated, missing, or out-of-order data

- Structured work-order, operation, defect, and lot data

- Production quantity, defect-rate, and operation-time aggregation

- CSV export and APIs for external analytics



Broadway may be used as a high-throughput ingestion layer for equipment data.

It handles concurrency, back-pressure, batching, retries, and failure

isolation; it is not the analytics engine itself.



## Architecture



```text

Browser / Tablet

       │

       ▼

   Web / API ── MES Core ── PostgreSQL

       │

       ├── Background Jobs

       ├── Audit Log

       └── Event Outbox



PLC / Sensor / Equipment

       │

       ▼

 Edge Connector ── HTTP / MQTT ── Ingestion

                                      │

                         Broadway Extension (optional)

                                      │

                         Telemetry Store (optional)

```



Equipment events do not automatically become confirmed production results.

They are stored as candidates and matched against work orders, operations,

operators, and lots before confirmation.



## AI Principles



AI is not an autonomous production-data operator.



| Stage | Allowed behavior |

|---|---|

| Read | Query authorized production and lot context |

| Suggest | Propose anomalies, patterns, and checks |

| Approve | Let an authorized person review the proposal |

| Execute | Run only explicit approved actions with an audit record |



AI accesses data through a permission-aware Context API rather than querying

the database directly.



## Project Status



| Area | Status |

|---|---|

| Vision and MVP | Documented |

| Domain model | Draft complete |

| System and AI architecture | Draft complete |

| Manufacturer adoption guides | Draft complete |

| Application scaffolding | In progress |

| MES Core | Early implementation |

| Lot traceability | Planned |

| Data collection and analytics | Planned |

| Multilingual UI | Planned |

| Production release | Not released |



## Getting Started



There is no official runnable release yet. Start by reading:



1. [Vision](docs/vision.md)

2. [MVP Scope](docs/mvp-scope.md)

3. [Domain Model](docs/domain-model.md)

4. [System Architecture](docs/system-architecture.md)

5. [Contributing Guide](CONTRIBUTING.md)



Verified installation instructions and Docker Compose files will be added

after the first runnable scaffold is complete.



Factory technology owners can use the public

[AI adoption validation prompt](https://openmeskorea.org/#adoption-prompt)

with an LLM. The site also publishes

[`llms.txt`](https://openmeskorea.org/llms.txt).



## Roadmap



- [x] Phase 0: Vision, scope, and architecture

- [ ] Phase 1: MES Core

- [ ] Phase 2: Lot Traceability

- [ ] Phase 3: Shop-floor UX

- [ ] Phase 4: Data Collection and Basic Analytics

- [ ] Phase 5: AI Read-only and Decision Support

- [ ] Phase 6: Integrations and Extensions



## Contributing



The project currently needs:



- Reviews of manufacturing terminology and exception workflows

- Elixir/Phoenix and PostgreSQL implementation

- Shop-floor tablet UX

- Experience with lots, quality, equipment, and ERP integration

- Translation into English, Japanese, Chinese, and additional languages

- Testing of security, performance, and disaster recovery



Read [CONTRIBUTING.md](CONTRIBUTING.md) before contributing. Large features

should first be evaluated as either core behavior or an extension.



## Non-goals



- Replacing all ERP accounting, HR, purchasing, and sales functions

- Bundling every industry rule and PLC driver into the core

- Shipping advanced APS and full automation in the first release

- Allowing AI to modify production data without approval

- Presenting unimplemented features as production-ready



## License



This project is distributed under the [MIT License](LICENSE).



Copyright (c) 2026 Baryon Labs



---






**Start small. Trace everything. Extend deliberately.**