Keep your AI. Let Coilmind handle TIA Portal.

How the two agents work together

Coilmind runs on your PC next to TIA Portal and can act as an agent for another AI. Your agent hands over a whole task, for example “list the SCL blocks in this project” or “implement this block from this spec”. Coilmind’s own agent carries it out in TIA and sends back the result.

Your agent cannot skip the checks. Every block change goes through the compile gate, the same as when you ask Coilmind directly.

How this differs from a raw MCP server for TIA Portal

There are MCP servers that expose the Openness API to a language model as a set of low-level tools: export this block, import this XML, compile. With those, your model has to know Openness, the XML format and the topology of a LAD network by itself, and it has to get each step right.

Coilmind takes the task as a whole. Your model decides what should be done, and a specialised agent does the TIA part, including the checks and the rollback when something does not build.

Your model plans, Coilmind’s agent does the TIA work

A strong general model is good at breaking a job into steps and at reviewing what came back. It usually does not know how to lay out a LAD network that TIA will accept, or how a particular instruction has to be called on your CPU.

Coilmind’s agent was built for that part. It works from the official Siemens documentation, knows how a LAD network is assembled, and keeps fixing a block until it compiles. So your agent plans the work and checks the results, and Coilmind does the writing, importing and compiling in TIA.

This also helps on a hard task where Coilmind’s built-in model would struggle with the planning. Your model takes over the thinking, and the work inside TIA stays with the agent that knows it.

Less noise in your agent’s context

The raw description of one block can run to many kilobytes of XML. If your agent pulled that for every block it looked at, its context would fill up long before the job was done. Through Coilmind it gets the result of the task, a list, a summary or a confirmation that the block compiled, and keeps its context for the actual work.

You can watch the conversation

The exchange between your agent and Coilmind’s agent is shown live in a separate chat inside Coilmind. At any moment you can open it and see which task your agent sent and what Coilmind is doing in the project about it. Every change it makes also goes into the project changelog, and blocks and screens are snapshotted before they are touched.

You don’t need your own agent

Coilmind works on its own, with its built-in chat and an AI model included in every plan. You can also point it at your own OpenAI-compatible key or a local model through Ollama. Connecting an external agent is an option for people who already have one in their workflow.

Why ChatGPT in a browser tab is not enough

Plain ChatGPT in the browser does not see your project, so it does not know your tags, your blocks or your CPU. It can only answer in text, which means it cannot hand you a LAD network, only a description of one. And it cannot compile anything, so you find out what is wrong when you paste it into TIA yourself.