topic Re: Using an LLM with ThingWorx - Latest Version (V2) Update! in ThingWorx Developers

Using an LLM with ThingWorx - Latest Version (V2) Update!

d_kessler — Fri, 10 Oct 2025 18:33:52 GMT

New Best Practice: Using LLMs with ThingWorx — Version 2.0 Now Available!

Everyone is talking about Generative AI, and Large Language Models (LLMs) continue to transform how organizations analyze and interact with data. Beyond chatbots and text generation, these models are increasingly being used to make sense of complex industrial data — from factory sensors to production analytics — helping teams make faster, smarter decisions.

ThingWorx, PTC’s Industrial IoT platform, enables secure connectivity to machines, systems, and databases across the enterprise. By linking ThingWorx with an LLM, you can create an AI Assistant for industrial operations that provides insights, answers natural-language questions, and supports data-driven problem solving — all while keeping your operational data secure.

🧩 What’s New in Version 2.0

This new release builds on the original ThingWorx + LLM accelerator and introduces major new capabilities designed to make LLM integrations more powerful, flexible, and enterprise-ready:

Integration with ThingWorx Analytics
The accelerator now connects directly with ThingWorx Analytics Server to interpret results from Signals, Profiles, Datasets, and Predictive Models. You can ask questions like:
“What factors most influence grinder failures?”
“How accurate is the predictive model trained on my dataset?”
The LLM analyzes the context and provides clear, actionable insights in natural language.
Router-Based Architecture
Version 2 introduces a “router” service that allows the LLM to intelligently select which ThingWorx services to execute based on the user’s question. This makes the system more efficient, cost-effective, and scalable — only the necessary data and context are sent to the LLM.
Dual-Tab Mashup Design
The new Insights Mashup includes two tabs:
- Data Insights (V1) for simple natural-language data queries
- Analytics Insights (V2) for interpreting ThingWorx Analytics jobs
  The design follows PTC’s corporate template for a modern, consistent experience.
Expanded Service Library & Security Enhancements
The Manager Thing now contains 17 modular services, improving transparency and maintainability.
This version also incorporates Azure OpenAI best practices for secure key management, data minimization, and compliance.

What You Get

The Accelerator includes:

A ThingWorx project file (PTCDTS_TWXLLM_V2.xml) that can be imported directly into ThingWorx 9.5 or above
An updated Best Practice document (V2) detailing setup, configuration, and validation steps
Sample data files (DEMO_BEANPRO_DATA.csv and metadata JSON) used to demonstrate Analytics Insights

Together, these resources show how to connect ThingWorx and an LLM using secure API calls — no external libraries required — and how to extend the architecture to your own factory data and analytics models.

Watch the New 21-Minute Walkthrough

To accompany this release, we’ve published a 21-minute video overview and live demo that walks through:

How ThingWorx and LLMs communicate
The new router-based architecture in Version 2
How Analytics Insights can interpret predictive model results
A full end-to-end demo showing natural-language queries in both tabs of the accelerator

Whether you’re a developer, system integrator, or operations engineer, this walkthrough shows how easily you can bring Generative AI into your ThingWorx applications.

Get Started

You can download everything — the Best Practice Document, Accelerator Project, and supporting demo data — from the official GitHub repository:
ThingWorx LLM Accelerator on GitHub

Import it into your ThingWorx environment, enter your Azure OpenAI credentials, and you’ll be up and running in minutes.

By combining ThingWorx’s industrial connectivity with the reasoning and language capabilities of LLMs, this accelerator helps organizations turn raw IoT and analytics data into real, explainable insights — safely, securely, and efficiently.

(view in My Videos)

Re: Using an LLM with ThingWorx - Latest Version (V2) Update!

smainente — Tue, 21 Oct 2025 10:00:20 GMT

@d_kessler - Can you check the content of the V2.0 release in github ? I see only v1 resources.

Re: Using an LLM with ThingWorx - Latest Version (V2) Update!

JK_10744682 — Tue, 04 Nov 2025 19:41:18 GMT

Is there information on hooking this up to my own data in my TWX environment?

Re: Using an LLM with ThingWorx - Latest Version (V2) Update!

JK_10744682 — Wed, 05 Nov 2025 15:09:38 GMT

I am also having issues with being unable to download the BEANPRO csv and meta data from github. Says something went from when downloading from the given link.

Re: Using an LLM with ThingWorx - Latest Version (V2) Update!

d_kessler — Wed, 05 Nov 2025 17:44:32 GMT

Hello - I've updated the structure of the GitHub repository so that now it has the V2.0 files in a new folder "v2.0_current".

Re: Using an LLM with ThingWorx - Latest Version (V2) Update!

d_kessler — Wed, 05 Nov 2025 18:01:58 GMT

How to hook it up (quick steps):

Point to your source: Use ThingWorx connections you already have (Value Streams, SQL, historians/OPC UA, REST). You can keep the demo Data Table or swap it for your own service that returns an InfoTable.
V1 (Data Insights):
- Replace the demo query in ProcessData with a call to your data service (or keep Data Table and load your data).
- Update the column list/context in CreateDataPrompt so the LLM generates valid queries/aggregations for your schema.
- Data never goes to the LLM—only the query + params do.
V2 (Analytics Insights):
- Create a ThingWorx Analytics Dataset from your data, then run Signals / Profiles / Model jobs.
- The existing V2 services will read those jobs and the router will select what’s needed to answer questions.
Credentials: Set GPTUrl, GPTAuth, and your Analytics server name on PTCDTS.TWXLLM.Manager.
Security best practice: Prefer Azure OpenAI; minimize/strip sensitive fields in prompts.

Full, step-by-step details are in the Best Practice V2 doc and the GitHub project readme.

Re: Using an LLM with ThingWorx - Latest Version (V2) Update!

d_kessler — Wed, 05 Nov 2025 18:06:42 GMT

Hi - please try downloading the file using the button highlighted in the attached picture.

Re: Using an LLM with ThingWorx - Latest Version (V2) Update!

nmutter — Fri, 21 Nov 2025 11:54:48 GMT

Related: We created an PoC with MCP for TWX which we can use for a chatbot with TWX data: Blog entry: https://blog.doubleslash.de/en/iot-and-connected-things/llm-based-agent-for-thingworx-architecture-and-implementation-with-mcp/ MCP server is in github if you want to try it yourself!

We really like the idea, which needs some more detailing on how to use it in production. But PoC is really fast - fascinating stuff asking the LLM something and it interacts with ThingWorx (read or write!)

It cannot do data analytics like the approach in the thread - but can be used for gernal information / control. (Or for sure extended with analytic models)

(PTC will release a MCP server themselves in the next months)