LADX
Start free
← Products

Chat

Available now

Ask for logic. Get logic that compiles.

Describe what the machine should do and watch the rungs appear on the canvas. Every answer passes a real compiler before it reaches you, and the failures go back to the model rather than to you.

Open Chat
The loop that makes it usableModel writesany providerCompiler answerspass / failYou read itonly if it passedfailures go back, with the errorsthis is what makes a free model good enough
Generation is a loop rather than a single shot. That is what makes a small free model good enough to be useful.

In short

Can AI write PLC ladder logic that actually works?

It can, if something checks it before you see it. LADX generates ladder from a description and then runs a real IEC 61131-3 compiler and the editor's own validator over the result, sending the errors back to the model rather than to you. What arrives on the canvas has already been refused once if it was wrong.

The problem

General AI assistants write plausible PLC code. Plausible is not the bar when the code runs a machine, and reading it carefully enough to be sure takes about as long as writing it would have.

What it does

  • It writes onto the canvas, not into a chat bubble

    Answers arrive as rungs in the Ladder editor: editable, runnable, and yours. Select a rung and ask what it does, or ask for a permissive to be added to it.

  • Validated before display

    matiec compiles it, iec-checker analyses it, a PLCopen schema pass checks the structure. Anything that fails goes back to the model with the errors attached, and you never see the attempts.

  • Any model, your key

    OpenRouter by default, including its free models, plus Anthropic, OpenAI, or any OpenAI-compatible endpoint including a local Ollama. Pick a model per conversation or leave it on Auto.

  • It knows the project it is in

    Tag names, existing routines, the target platform, and the design basis you filled in. Answers use the tags you already have rather than inventing new ones.

What it doesn't do

Listed because you would find out anyway, and finding out later is worse.

  • It drafts. Whether the interlock is sufficient for the hazard is an engineering judgement and stays with you.
  • A compiler proves code is valid, not that it is correct. Simulation helps; it is not a safety case.
  • Free models are rate-limited by the provider, not by us. The queue you occasionally hit is theirs.

Questions

Why does ChatGPT get ladder logic wrong?
Because nothing checks it. A model asked for a seal-in will often write a latch leg containing only the motor contact, which latches the motor on and leaves the stop button doing nothing. It looks right and it is a defect. The fix is not a better prompt, it is a compiler in the loop.
Which AI models can LADX use?
Any of them, through your own key: OpenRouter, Anthropic, OpenAI, or any OpenAI compatible endpoint. Because generated code is validated before it is shown, a small free model is usually good enough, which is the practical point of validating rather than trusting.
Does my code get sent anywhere?
The prompt and the program go to the provider whose key you connected, and nowhere else. LADX stores no shared key and trains on nothing. The desktop build makes no outbound call at all and runs its model locally.