Artoo - AI Agent
The Artoo modernisation turns a helpful chatbot into a capability platform: a graph-aware, state-driven agent that converses, reasons and acts. Backed by MCP and aligned with the wider OPB vision, Artoo will unlock measurable productivity gains for manufacturing customers while opening a sustainable partner ecosystem.
Vision & Objectives
Artoo will evolve from a helpful in-app chatbot into a domain-aware, action-oriented AI agent that understands manufacturing context, reasons over live plant data, and safely executes end-to-end workflows.
Conversational first – natural language becomes the primary UI for daily maintenance, analytics, and training.
Agentic intelligence – replace simple LLM chatbot with a state-based, goal-driven architecture able to plan, delegate, and verify multi-step tasks.
Open ecosystem – the new Model Context Protocol (MCP) exposes secure hooks so both internal micro-services and partner-built agents can co-operate inside P4.
Continuous learning – leverage a factory knowledge graph that captures system capabilities, factory assets, process and historical telemetry to improve answers and recommendations over time.
Problem Statement
The current chatbot answers FAQs and runs canned SQL reports, but cannot reason across tables, sensor feeds and work-orders.
Business logic lives in bespoke Python scripts; maintenance of intents, prompts and SQL snippets is brittle and opaque.
External consultants cannot safely extend Artoo without touching core code, hampering ecosystem growth.
Solution Overview
Layer | Responsibility | Key Technologies |
|---|---|---|
Conversation Orchestrator | Intent detection, dialogue state, hand-off to agents | Llama/OpenAI/Anthropic, RAG, LangChain/LangGraph |
Agent Runtime | Goal decomposition, tool-use, memory | Open AI Function-calling + custom State Machine Manager |
Knowledge & Context | Unified plant knowledge graph + vector store for unstructured docs | Neo4j, pgvector |
Action Toolkit | Secure, declarative wrappers around P4 APIs & SQL | FastAPI, Pydantic contracts |
Model Context Protocol (MCP) | Typed messages that describe context, permissions and expected artefacts | gRPC / protobuf |
MCP is shared with the Object Process Builder (OPB) initiative, ensuring every automated process—whether diagram-driven or AI-planned—speaks the same language across services and partners.
Key Capabilities (MVP → Target)
Capability | MVP (Q4 2025) | Target (2026) |
|---|---|---|
Multilingual Q&A | EN/CZ docs RAG | 10+ languages, voice |
Graph-aware reasoning | Static KG snapshot | Real-time KG sync with MES/APS |
Action execution | Create / update Maintenance Order, run parameterised reports | Any OPB node, cross-app orchestration |
Extensibility | Internal agents via MCP | Third-party agent marketplace |
Governance | Basic prompt logging | Full audit trail, role-based guardrails |
Architecture Evolution
Refactor Core (Q3 2025):
Replace monolithic Flask bot with micro-service Conversation Orchestrator.
Introduce vector search and first cut of knowledge graph.
Agent Platform (Q1 2026):
Deploy State Machine Manager with tool-usage limits & human-in-the-loop fallback.
Publish MCP v1 for internal services.
Ecosystem Enablement (H2 2026):
Harden MCP, add billing & throttling.
Release SDK and certification programme for partners.
Roadmap & Milestones
Quarter | Milestone | Outcome |
|---|---|---|
Q3 2025 | Tech Preview | Maintenance team pilots conversational work-order creation. |
Q4 2025 | MVP Launch | Artoo handles 30 % of help-desk queries; first KG-backed insights live. |
Q2 2026 | Consultant Beta | Partners builds custom agents capable of interacting with P4 by using MCP. |
H2 2026 | General Availability | Marketplace opens; >50 % routine tasks automated through Artoo. |
Business Value
35 % reduction in technician search time for documentation and historical fixes.
20 % fewer unplanned stops via proactive recommendations drawn from KG correlations.
New ARR stream – partner agents listed in marketplace on revenue-share model.
Differentiation – first AI agent in heavy-industry MES/APS that can both answer and act.
Stakeholder Impact
Stakeholder | Benefit |
|---|---|
Factory Management | Faster incident resolution, data-driven decisions, lower MTTR. |
Consultants & ISVs | Low-code entry point to offer domain expertise as plug-in agents. |
Sales & Marketing | Compelling demo of “AI that does”, shortening sales cycles. |
Internal Engineering | Clear boundaries; business logic moves from scripts to declarative tools. |