How We Securely Connect Your Plant Data to AI Tools
If you’ve tried asking an AI tool about your plant, you’ve probably done it the manual way. You export a CSV, paste it in and ask the question.
It works, but it can be slow and it leaves copies of your data everywhere. Export too little and you miss the thing that mattered. Export too much and you spend time cleaning it up before you can ask anything.
The Industrial App Store AI Connector sits between your AI tool and your plant data, so the export step disappears. It’s built on MCP, the Model Context Protocol. MCP is an open standard, so the connector works with tools you already have rather than a new one you need to learn. At the moment that means Claude and VS Code.
How it connects
The AI Connector reaches your data the same way every other app on the Industrial App Store does. A lightweight piece of software called App Store Connect sits on your business network with read-only access to your historians and alarm and event logs. It only makes outgoing connections over HTTPS, so nothing needs to connect into your network, and you decide which data sources are shared and who can see them.
When you ask a question, the AI tool calls the connector. The connector requests the data through the Industrial App Store, App Store Connect retrieves it from your historian, and the values come back to the AI tool to work with. If your site already has App Store Connect installed, all you need is a subscription to the connector.
With a cloud AI tool, the data the connector retrieves is sent to the AI provider to answer your question, just as it would be if you pasted it in. The difference is that only the data the question needs is sent, and no copies are left sitting on anyone’s laptop.
Plant data for people who don’t know the tag names
Getting data out of a historian has usually meant knowing exactly which tags to ask for. That puts a lot of useful information out of reach for anyone who doesn’t work with the historian every day.
With the AI Connector, you can ask about equipment in plain language, such as “the inlet valve on the second separator”, and the AI tool works out which tags it needs.
Note: we recommend checking each answer before acting on it, particularly where tag names or descriptions are inconsistent. Most AI tools will show you which tags were used, which makes that check quick. Keeping a person in the loop is the right way to work with any AI tool, and it means you always know what’s being asked of your plant data and how it’s being used.
Being able to ask in plain language opens the plant data up to people who couldn’t easily reach it before. A manager, an analyst or an engineer from another discipline can ask a question without first needing someone to find the tags for them.
Where it fits
The AI model is the same one you’d use anywhere else. The difference is in the work around the question. The tool finds the data source, pulls the right tags and retrieves the historical values itself. It works from the full history rather than whichever slice you remembered to export, and when a follow-up question needs different data, it fetches that too.
It works alongside your purpose-built tools rather than replacing them. If you need to monitor something continuously, an app designed for that job is still the right choice. The AI Connector is for the questions that don’t have a dedicated app yet.
Running it fully on site
Some sites can’t send data to cloud AI at all. For them, we install all the required software on site, so the system doesn’t need to connect to the internet and the data never leaves your network.
The cost works differently too. There’s an upfront investment in the hardware and infrastructure to run AI models in your own offices. Once that’s in place, you own and control the system completely. There are no per-use charges and no usage limits set by a provider, so you can keep it running all the time without worrying about hitting a cap partway through a piece of work.
Capacity depends on the hardware you choose, so you can start small and add more as your use grows. Local models aren’t as powerful as the largest cloud models, but for many everyday questions about your own data they do the job well. And within your capacity, the system is yours to use as much as you like.
Being offline with no usage limits makes some uses of AI practical that weren’t before. Work that would be too sensitive to send to a cloud provider, or too costly to run continuously at cloud prices, becomes something you can try.
For sites that can use the cloud, running AI on site doesn’t mean giving it up entirely. If you want, the same system can route some questions to a cloud model, such as harder questions that need a more powerful model, while everything else stays local. You decide what goes out and when, it all passes through one place you control, and you can change provider without changing how anyone works.
FreshAIR, our on-site AI system
This is how FreshAIR, our on-site AI system, works. The software and the AI models run on your own hardware and network, with cloud models available if you choose them. As well as plant data through the AI Connector, it works across your documents, mail, scans, drawings and spreadsheets, with every answer traced back to its source. Owning the system doesn’t mean being stuck with an old model either. As better open models are released, FreshAIR can be updated to use them, as long as they fit the hardware you have.
We use it ourselves every day, for writing reports, triaging incoming mail and turning scanned manuals and drawings into something searchable. We’ll cover it in more detail in a future post, including how it’s built, what it does well and where its limits are.
If you’ve been meaning to try AI on your plant data and the export step is what’s been holding you back, we’ve removed that step. Send us a message or get in touch at info@intelligentplant.com.
Comments