A decision intelligence platform for continuous industrial operations that automates operational decisions via structured multi-agent teams (ORPA cycle), overlaying existing SCADA/ERP systems with explainability and bounded autonomy.
Project Overview#
XMPro MAGS is a multi-agent generative system designed for industrial environments, built upon Stanford University's Generative Agents research (Park et al., 2023). The system employs non-disruptive integration, adding a decision intelligence layer atop existing historians, SCADA, digital twins, and ERP systems.
Core Design Philosophy#
- Non-intrusive Integration: Does not replace existing systems; adds an intelligent decision layer
- Separation of Duties: Agents proposing actions are separate from agents approving actions
- Bounded Autonomy: Agents only operate within engineer-approved boundaries; exceedances escalate to humans
- Explainability: Every decision includes documented reasoning that operators can accept or challenge
ORPA Decision Cycle#
Agents employ an Observe-Reflect-Plan-Act architecture:
- Observe: Monitor operational data via DataStream
- Reflect: Learn from historical data and patterns
- Plan: Formulate multi-step action plans
- Act: Execute decisions and coordinate with other agents
Dual-Layer Intelligence Architecture#
- Business Process Intelligence Layer (~90%): Decision-making, planning, memory, optimization with 15 core capabilities
- LLM Tooling Layer (~10%): Natural language processing for communication and explanation
Agent Types#
| Type | Description |
|---|---|
| Content Agents | Content processing agents |
| Cognitive Agents | Cognitive agents |
| Hybrid Cognitive Agents | Hybrid cognitive agents |
Primary Use Cases#
- Predictive Maintenance
- Process Optimization
- Quality Management
- Safety-Critical Operations
- Root Cause Analysis
- Compliance Management
Example Team Configurations#
- Advanced Predictive Maintenance Team
- Antibiotic Fermentation Optimization Team
- Expert OEE Optimizer Team
Business Value Metrics#
- 30% downtime reduction
- 25% process optimization
- 80% faster implementation
- 15%+ productivity improvement
Configuration#
Repository Structure#
/docs — All documentation (28 topic areas)
/src — Agent configuration templates and examples
/case-studies — Real-world implementation case studies
/research — XMPro research papers
Configuration Files#
- Team Manifests (
src/team_manifests/): JSON files defining team structure, goals, communication protocols, decision workflows, operational constraints, escalation strategies - Agent Profiles (
src/agent_profiles/): JSON files specifying agent characteristics, capabilities, roles, skills, rules, model specs, memory parameters
Deployment#
- Docker containerized deployment supported
- Database initialization scripts:
constraints.cypher,library_prompts.cypher,library_tools.cypher,system_options.cypher
Technical Architecture#
- Memory System: Vector database-based agent memory
- Monitoring: OpenTelemetry tracing integration
- Integration: DataStream integration, tool orchestration
- Decision Orchestration: Inter-agent coordination and consensus mechanisms
Important Notes#
⚠️ Key Information:
- MAGS platform core runtime is commercial closed-source software
- GitHub repository provides documentation and configuration examples only
- Contact support@xmpro.com for platform licensing
- Specific programming languages and frameworks not explicitly stated in repository
- Cloud deployment options to be confirmed