Offline AI for PLC programming means the language model and your programs both stay on the engineering workstation, with no call to a cloud service. In practice that is a local model, run with a general tool such as Ollama, so prompts, programs and drawings stay on the computer.
That matters on sites where the network policy says customer code does not go to the cloud, and on isolated OT networks where the engineering laptop has no internet route at all. Water and power utilities, pharmaceutical plants, defence suppliers and many food manufacturers fall into one or both groups.
Why cloud AI is a hard sell on OT networks
A PLC program encodes how a process runs: sequences, setpoints, interlocks and the know-how a manufacturer has built over years. Sending it to an external AI service is a data transfer that many security policies forbid outright, whatever the provider's terms say about training.
Guidance for OT security points the same way. NIST SP 800-82 Rev. 3, published in September 2023, covers how to secure industrial control systems while respecting their performance, reliability and safety needs, and the ISA/IEC 62443 series is built on zones and conduits that limit what may cross from one network to another. A tool that needs a live internet connection to work creates a conduit somebody has to justify. A tool that needs none does not.
How offline AI for PLC programming works
| Item | Local model on an isolated workstation | LADX (web tool at ladx.ai) |
|---|---|---|
| Where it runs | On the engineering workstation | In your browser, with AI requests sent from ladx.ai's server |
| AI inference | A model on the same machine | Free models through LADX's shared account, or your own provider key |
| Network use | None after set-up, if you keep updates off | Online |
| Local Ollama model | Yes, that is the point | No; the server cannot reach localhost on your machine |
| Suited to | Controlled or client-confidential programs | Non-sensitive drafting, learning and checking on a connected machine |
Ollama is an open tool for running open AI models on your own machine. Model weights alone are gigabytes, so you install the tool and pull a model yourself on a connected staging machine. Note that Ollama can also run models in its own cloud; on an isolated site, make sure only local models are configured.
A step-by-step set-up for an isolated workstation
- Agree it with the site's OT security owner first. Describe the software, the model and the transfer route. Get it into the approved software list before anything is installed.
- Prepare on a staging machine. On a connected PC, download the installer, check its hash against the published value where one is given, and pull the model you intend to use.
- Transfer through your approved media route. Use the site's scanning and logging procedure for removable media, as you would for a firmware update.
- Expect an operating system warning. Unsigned installers trigger one. Verifying the hash is your evidence that the file is the one published.
- Leave automatic updates off. On an isolated machine, bring new versions in through the same controlled route.
- Put prompts, drafts and accepted code under backup. Keep them in a project folder, so they can be archived with the site's other control system records.
What is Ollama? Running Local LLMs Made Simple
IBM Technology
This video is hosted on YouTube. Playing it loads content from Google, which may set cookies.
Watch on YouTubeWill a local model be good enough?
Local models are smaller than the largest cloud models, and you should expect that. A small model needs checking: compile its output with the vendor's tools, review it and test it, as we explain in can AI write PLC code that works?
Size the workstation for the model you choose rather than the other way round. Try two or three models on the staging machine with your own typical tasks: a seal-in with permissives, a step sequence, an alarm block. Keep the one that gives clean drafts in an acceptable time on the hardware the site actually issues.
Where LADX fits, and where it does not
LADX is web only and free. It has no desktop app, no "LADX Studio" download, no offline mode and no installer, and it cannot use a local Ollama model, so it does not belong inside an air gap. "LADX Studio" is now only the name of the signed-in web workspace.
It suits non-sensitive drafting, learning and checking on a connected machine. The ladder editor and simulator, converters, HMI builder and panel CAD run in the browser with no account, and chat needs a free account. Generated ladder is validated and given a seal-in check, Structured Text is compiled by the LADV compiler, and problems are listed beside the answer. Code that passes can be run on a Wartens LADV virtual PLC at ladx.ai/vplc, which is not real time, not safety rated and cannot control machinery. On the free tier, the company running the free model may keep prompts and answers, so keep site data out.
What offline does not change
- The model still never writes to a controller. Download stays a deliberate act in the vendor's software. LADX likewise never connects to or downloads to a real controller.
- Review and test still apply. Run drafts through the vendor's simulator with fault and restart cases, then through your normal FAT and SAT.
- Your OT procedures still govern change. Treat AI-assisted edits like any other modification under management of change.
For broader context on segmenting control networks, our guide to IEC 62443 for plant managers is a practical starting point, and the UK NCSC operational technology collection is a useful free reference.
Next step
If you run a regulated or isolated site and want AI assistance without a cloud dependency, read the LADX product brief for what the web tool does and does not cover and the industrial AI practice page, then contact us to talk through the deployment.
