AI Integration

FrameworX provides practical AI capabilities for industrial automation — from AI-assisted solution building to runtime data queries, machine learning deployment, and portable AI skills for Claude and other AI agents.

FrameworX AI Capabilities

AI Designer: What's Inside

AI Designer is not a single tool — it's a comprehensive package of components that work together to make AI an expert FrameworX engineer.

Live IDE co-pilot (DesignerMCP) — Claude connects directly to the running Designer and builds solutions alongside the engineer. Tools cover the full solution lifecycle — tags, displays, alarms, devices, protocols, historian, scripts, and runtime. Every tool call produces immediate visual changes the engineer watches in real time. Both open_solution and create_solution accept an optional from_workspace=<path> parameter that imports a ConsoleMCP-authored workspace folder into the solution after opening.

Offline file engineering (ConsoleMCP) — Claude Code generates FrameworX JSON configuration files without a running Designer, using workspace vocabulary (list_workspaces, open_workspace, create_workspace, get_workspace_info). A dedicated create_solution_file tool invokes SolutionCreator.exe headlessly to compile the workspace into a deployable .dbsln. Ideal for building solutions from specifications, analyzing existing projects, and batch engineering across multiple solutions. The engineer imports the files into Designer for validation and deployment.

Claude Skill — A portable SKILL.md file that loads into Claude (and other AI agents) at the start of every session. Provides baseline FrameworX knowledge, progressive build discipline, and MCP setup guidance. The behavioral foundation that makes every AI session productive from the first response.

Skills Library — Build playbooks that AI Designer loads during construction via search_docs. Step-by-step recipes for complex patterns — new solution builds, display construction, alarm pipelines, MQTT/SparkplugB integration, edge ML deployment. The AI loads the right playbook at the right time.

Extensibility — Custom MCP tool plugins (DesignerMCPCustom*.dll) extend AI Designer with company-specific tools. The Skills Library is open for custom skill authoring. The Claude Skill follows the open Agent Skills standard and works across Claude, GitHub Copilot, Cursor, and any compatible agent.

What makes AI Designer different

Most MCP integrations are thin wrappers — 3-5 stateless tools that read and write through an API. The AI has no memory of the platform, no understanding of relationships between objects, no guardrails. Every session starts from zero.

FrameworX AI Designer is architecturally different:

• Progressive knowledge delivery. Architecture concepts on connection, module schemas on first access, field-level guidance on each schema fetch, build playbooks on demand. The AI receives exactly what it needs at the moment it needs it.
• AI/human object ownership. The MCP Category system creates a clear collaboration boundary — AI-created objects are tagged and modifiable by AI, engineer-created objects are read-only to AI unless explicitly unlocked.
• Session-aware behavior conditioning. The Claude Skill loads behavioral rules before MCP tools connect. The MCP context reinforces them throughout the session. The AI doesn't just have tools — it has trained instincts for how to use them.
• Live visual feedback. Every tool call produces immediate changes in Designer. Orange border and "AI Designer" badge make the AI connection visible to the engineer.
• Protocol intelligence. Fuzzy matching finds the right driver by vendor name. Pre-built Wizard Symbols for industrial equipment. Domain knowledge embedded in tool responses.

The closest comparisons in AI-assisted development are tools like Cursor and Windsurf — but those are AI-native editors built from scratch. FrameworX is a mature industrial platform with 30+ years of domain expertise and 5,000+ deployments that has achieved the same depth of AI integration. The AI co-pilots a proven production platform, not a prototype.

What to Expect

AI Designer produces the best results when you work with it incrementally, following the solution pillars in order: tags, devices, alarms, historian, displays, scripts. Building everything from a single prompt is possible, but complex solutions built that way typically need more iteration and review.

How to approach a new build:

• Structure your prompts with three parts: a role (what Claude is), context (your solution and goals), and instructions (what to do). This three-part structure consistently produces better first results. See the Best Practices page for details and examples.
• Follow the pillars in order and confirm each layer is working before moving to the next.
• Define what you know: protocol, folder paths, data types, alarm setpoints. For anything you are unsure about, ask Claude to suggest options rather than guessing yourself.
• If something doesn't work as expected, describe the problem to Claude directly. Most errors resolve in one to two follow-up messages without needing to restart.

Current limitations to be aware of:

• Display graphics: AI-generated displays are structurally correct, but the visual layout, element sizing, and alignment typically need manual refinement in Designer.
• Image placement on screens: AI does not reliably place images on displays. Do this manually.

→ See AI Designer in Action

→ See AI Designer Connector

→ See MCP and Claude Setup

AI Tutor: Learn FrameworX Interactively

A structured curriculum delivered through Claude in your Designer MCP session. 45 lessons across three tiers cover everything from your first tag to enterprise multi-site architecture. The AI Tutor turns documentation into interactive, hands-on learning — every lesson builds working features in a real solution.

What makes it different from documentation:

• Hands-on by default. Every lesson builds something in your solution that you can run and verify — not slides, not videos, not throwaway exercises.
• Adaptive depth. Ask "explain that simpler" or "go deeper" — the lesson follows your needs. The AI is your tutor, not a script.
• Progress that survives upgrades. Lesson completion is tracked per user. New product versions add new lessons without losing your progress.
• No separate setup. If you have AI Designer, you have the Tutor. Just ask Claude to start a lesson.

Three tiers, 45 lessons:

• Essentials (15 lessons) — From "what is FrameworX" to a working mini SCADA. First-run competence with tags, devices, alarms, historian, displays, and your first script.
• Intermediate (15 lessons) — All platform modules in production context. UserType composition, advanced alarms, historian compression, datasets, reports, scripts, symbol authoring. Ends with a brewery integration capstone.
• Advanced (15 lessons) — Architecture, deployment, integration, AI-assisted engineering. Execution domains, data servers, Sparkplug B, Python/ML.NET, distributed runtime, CI/CD. Ends with an enterprise multi-site SCADA capstone.

How to start: In any Designer MCP session, ask Claude:

• "Teach me FrameworX from the beginning" — starts at E1
• "Start lesson E5" — jumps to a specific lesson
• "I want to learn about alarms" — topic-driven entry, the Tutor picks the right lesson
• "Where am I in the curriculum?" — resumes where you left off

The Tutor recognizes natural learning intent. Any FrameworX MCP session can become a lesson on demand. Lessons are also discovered automatically when a new user opens Designer for the first time.

AI Runtime: Query Live Data

Connect AI models like Claude to your running solution. Claude, Copilot, and Cursor access tags, alarms, and historian directly — for dashboards, troubleshooting, and automated reports. Visual feedback: orange border and badge when AI is connected.

What AI can do:

• Read tag values from the UNS
• Query alarms and historian
• Create custom queries via Script Classes

→ See AI Runtime Connector

ML Integration: Machine Learning Inside Solutions

Deploy trained models for consistent, repeatable results. Two paths: ML.NET 4.0 in C# for deterministic inference, and Python 3.7+ for the full analytics ecosystem.

Use cases:

• Anomaly detection
• Predictive maintenance
• Quality prediction

→ See AI ML Integration Connector 

→ See Python and .NET Integration

Ask FrameworX Docs

ChatGPT trained on FrameworX documentation.

Click here → AI Ask Docs, or use the shortcut on the left sidebar.

Quick Links

In this section...