Shopfloor Copilot — AI-Powered On-Premise Manufacturing MES

AI-Powered Manufacturing Execution System (MES)

Version 1.0 Live Demo & Pilot Evaluation Connect · Contextualize · Analyze
Launch Guest Demo → Request Demo Credentials Discover Features ↓

By Machina Intelligence. Currently in live demo and pilot evaluation. The public hosted instance uses simulated plant data and a cloud LLM for speed. On-premise pilots run locally with Docker Compose and Ollama.

scroll↓

Platform Overview

From Monitoring to
Intelligent Decision Support

Shopfloor Copilot is an AI-enhanced MES designed to modernize industrial monitoring through intelligent decision support and real-time data analysis. It integrates OPC UA protocol connectivity with a semantic engine to transform raw machine signals into actionable insights — OEE metrics, standardized loss classifications, and AI-grounded diagnostics with source-grounded answers for human review.

30+
Manufacturing KPIs
11
Simulated Lines
200+
Stations Supported
100%
Local-First Pilot Architecture
🔌
OPC UA Connectivity
Native OPC UA integration with semantic engine. Raw machine signals become OEE metrics and standardized loss classifications automatically.
🤖
RAG-Grounded AI Diagnostics
Evidence-based troubleshooting using Retrieval-Augmented Generation. Answers grounded in your SOPs and Work Instructions — with citations for human review.
🤝
Digital Collaboration
Structured digital shift handovers, team chat, operator logbooks, and NCR management — replacing paper logs and informal phone calls.
🔒
Local-First Privacy
Built to run entirely on-premise: local LLMs via Ollama and a ChromaDB vector store, so production data never leaves your server. This public demo runs diagnostics on a cloud model to keep responses fast.

Platform Capabilities

Everything Your Shopfloor Needs

12 integrated modules. Select a capability to reach guest demo access (read-only, one-hour session), or request a guided evaluation. From real-time OEE monitoring to AI-powered Q&A, semantic signal mapping, and predictive maintenance.

📊
Real-Time OEE Dashboard
Live Availability, Performance, and Quality metrics per line and station. Color-coded status tiles and shift downtime events.
Open Dashboard →
🧠
AI Diagnostics & Copilot
AI Copilot analyses station data in real-time, providing root-cause analysis with checklist steps grounded in your documentation.
Open AI Diagnostics →
💬
Operator Q&A Knowledge Assistant
Ask natural-language questions. Answers from your WIs, SOPs, and Safety Guides via RAG + ChromaDB. Source citations and relevance scores included.
Ask a Question →
🔌
OPC UA Explorer
Browse live OPC UA node trees, read real-time values, and add nodes to a persistent watchlist. Direct connection to your PLC servers.
Explore Nodes →
⚡
Semantic Signal Mapping
YAML-driven engine maps raw OPC signals to standardized loss categories — availability, performance, quality. No hardcoding.
View Signals →
🏭
OPC Studio KPI Engine
30+ manufacturing KPIs via AST formula parser: Availability Rate, MTBF, MTTR, Throughput Efficiency, Energy per Unit — live.
View KPIs →
🔴
Andon Board
Color-coded station × line grid. Green ≥85% OEE, red <50%. One-click alert acknowledgment. Kiosk mode for shop floor wall displays.
Open Andon →
🔧
Predictive Maintenance
Equipment health scores (0–100), failure probability, and predicted failure dates. 48-hour Prophet time-series forecast per asset.
View Maintenance →
📈
Advanced Analytics
OEE trends, bottleneck detection (TOC), root cause Pareto, comparative cross-line benchmarking, and defect risk predictions.
Open Analytics →
🤝
Digital Shift Handover
Structured handover with auto-populated open issues, operator notes, and OEE summaries. Email delivery to incoming shift.
Open Handover →
⚠️
Violations Management
Signal threshold violations with SLA timers per severity. Acknowledgment workflow, comment trail, and per-station compliance stats.
View Violations →
🌐
Visual Line Designer
BPMN drag-and-drop production line editor. Define station types, cycle times, and critical paths. Syncs to live plant model instantly.
Open Designer →

System Architecture

Tech Stack & Infrastructure

Modern 3-tier containerised architecture. Local-first by design — deployable with no cloud dependency and no PII leaving your network. This public demo uses a cloud model for speed.

Frontend
🖥️ NiceGUI 2.5 ⚡ Vue.js / Quasar 🎨 Tailwind CSS 📱 PWA / Mobile
Backend & AI
🐍 Python 3.11 ⚡ FastAPI 🗄️ PostgreSQL 16 🧠 Ollama (Local LLM) 🔍 ChromaDB 🔗 OPC UA
Infrastructure
🐳 Docker 24.0+ 📦 Docker Compose 🔐 JWT Auth 🌐 Nginx Reverse Proxy 🔒 On-Premise Capable
🏠
Local-First Architecture
Deployable fully on-premise — local inference, no PII leaving your network, no cloud API calls. This public demo runs diagnostics on a cloud model so responses stay fast.

Connectivity

Works With Your Stack

Connects to PLCs, SCADA, ERP, MES, and CMMS systems through open protocols. No vendor lock-in.

🔗
OPC UA
Native client + semantic mapping
🤖
MCP
Model Context Protocol for AI agents
🏭
SAP / ERP
REST API integration layer
📊
SCADA
Via OPC UA gateway
🔧
CMMS
Maintenance ticket sync
📋
QMS
Quality & violation records
🧠
LLM / Ollama
Local AI inference, no cloud
🗃️
PostgreSQL
Time-series + relational data
🧩
Multi-Agent Intelligence
5 specialist AI agents — OEE Analyst, Maintenance Advisor, Quality Inspector, Shift Coordinator, and Knowledge Agent — route and synthesize answers to complex questions. The MCP interface connects manufacturing context and tools to MachinaOS for governed multi-agent orchestration.

Deployment

Server Requirements

Runs on standard hardware. Deploy on-premise or on a private VPS. Docker + Linux — no special cloud infrastructure required.

⚡ Minimal / Pilot
🖥️ 4+ CPU Cores
💾 16 GB RAM
📀 100 GB Storage
🐳 Docker / Linux

CPU-only inference works with quantized small models (3B–7B), with slower responses. An external API fallback is optional and sends prompts off-premise.

🏆 Recommended — Production
🖥️ 8+ CPU Cores
💾 32–64 GB RAM
📀 200 GB Storage
🐳 Docker / Linux

Recommend an NVIDIA GPU with 16–24 GB+ VRAM for low-latency, concurrent local Ollama inference across the 5 specialist agents and RAG pipeline. Size and validate for model, context length, and concurrency. Production readiness and scalability require pilot validation.