Yes, there is an LLM in it. But you don’t paste code into a chat window: it works inside the TIA project you have open.
An AI that writes ladder logic into your TIA Portal project, and only applies it if it compiles.
It reads the open project through Openness, writes the blocks, imports them and compiles. If the build fails, the change isn’t applied.
A clean compile doesn’t tell you the logic does what you meant, so Coilmind also tests it on the simulator: watch 0 passed → 2 / 2 pass →
Windows · TIA Portal V17–V21 · Openness API · 7 UI languages · PLCSIM Advanced V3–V6 (tests only, sold by Siemens) · svghmi.pro · tools for Siemens engineers
The 70-second tour
Four chapters, all on one project. It reads the logic, writes three networks that compile, runs them against the simulator and takes a snapshot before each change.
Jobs that eat the week
"Cloning the same block for 12 conveyors by hand, renaming every tag."
→ block cloning"Inherited a 15-year-old project full of STL nobody on the team can read anymore."
→ legacy STL → SCL"Reverse-engineering a block someone else wrote, network by network."
→ code analysis"Translating comments and tags into three languages for the client’s HMI."
→ translation"Editing WinCC Unified scripts in a window with zero IntelliSense."
→ WinCC editor"Copy-pasting XML into ChatGPT and back. It never sees the project, and it can’t change anything in it."
→ works in the open projectWhat Coilmind does
Each of these works on the project you have open in TIA Portal.
Answers questions about the whole project
Ask what an FB does, who writes to a tag, what’s inconsistent. The answer comes from your project, not from general PLC knowledge. It reads pasted screenshots too. Outside the chat there are project dashboards: call graph, architecture and health, instance usage, and which blocks are actually reusable.
Writes the documentation from the LAD and SCL
Point it at a block, a folder or the whole program. It reads the LAD and SCL, works out what the logic does (interlocks, sequences, which block writes to which tag) and writes it up: what each block is for, its interface, and a flowchart in Mermaid. You get a Markdown file that opens in any editor and can go to the client as it is.
Clones a block and renames the tags
“Clone FB Motor1 for the second drive, use M2_* tags.” It creates the copy, imports it and compiles, and keeps fixing it until it builds.
Twelve conveyors from one template, in one pass
Queue a template against a list. It creates every block, allocates the tags, walks the start addresses and gives you a report of what it made. Deterministic and repeatable, with a history of every run. For conveyors two to twelve it is the fastest route, and no model is involved.
Works through the steps itself
You ask in chat. Coilmind writes, imports, compiles and fixes its own errors, and takes the next step itself when it is obvious. Every block and screen it touches is snapshotted first, so a change you don’t like is one click back, and every action is listed in the project changelog.
Rewrites legacy STL as SCL or LAD
Legacy STL blocks get rewritten in chat. To prove the rewrite behaves like the original, run the original’s tests against it; that part needs PLCSIM Advanced. Anything that doesn’t translate cleanly is shown to you instead of guessed.
Works offline, up to 14 days at a time
Run a local model on the engineering PC itself, on site or anywhere off the network. Your code stays on the PC. You can work up to 14 days without a connection, then the licence needs a short refresh. Local models are weaker than the cloud ones, and how capable they are depends on your GPU.
Translation that keeps the PLC terminology
Inherited a German or Chinese project? It translates titles, comments and tag descriptions with the logic as context, so you don’t get machine pidgin back.
Block templates, no AI involved
Mark the tags and parameters that change, then type the new names. Fully deterministic, so you can use it where AI is off-limits.
Builds the HMI screen end to end
It creates screens, places items and changes any item’s properties: position, size, colours, text, the tag it shows. It creates the HMI tags, tag tables and groups behind it, sets dynamizations, writes the JavaScript for events and script dynamizations, and compiles the HMI. Scripts you write yourself get a proper editor: Monaco, with IntelliSense from the Siemens V21 definitions and project-wide search. It does not create HMI devices. Faceplate types are out of reach for any tool, because Openness cannot create them.
Tests stored with the project
Tests are declarative .test.json files: set inputs, run N cycles, assert outputs, and watch the run live. Coilmind can scaffold a suite from a spec and download to the simulator, never to real hardware. This needs PLCSIM Advanced, which Siemens sells separately.
Works as the TIA hands for Codex or Claude Code
Connect your own agent, such as Codex or Claude Code, to Coilmind over local MCP. Your agent plans and sets the task; Coilmind does the TIA part with what it knows about Openness, LAD networks and Siemens instructions, and with the same checks as always: dry-run validation, compile gate, snapshot. The conversation between the two agents is visible in Coilmind while it runs.
One session, nothing cut
Dry-run protection written into an existing pump block, then tested on PLCSIM until it passes: one session, 9 minutes 36 seconds of recording, nothing cut. There is one break of 36 minutes while PLCSIM Advanced was started, and the frame says so. Typing and generation waits are sped up. Nothing else is edited.
If you’ve seen AI demos that fall apart the moment you try them yourself, this is the one to watch.
Where it differs from the cloud assistants
It works like the cloud assistants, but it doesn’t expect you to be online all the time, to pay for a year up front or to upgrade to the newest TIA Portal.
With a local model, your code stays on the PC
Run a local model on the PC itself, on site or on an air-gapped network. The block templater needs no model at all. Local models are weaker than the cloud ones, and how far they go depends on your GPU budget.
Priced per licence, month to month
Pay by card, month to month. No annual lock-in and no org account. AI is included, with no tokens to buy.
Works with TIA Portal V17 to V21
On V17 or V18? Coilmind attaches to the project as it is, no forced upgrade to the latest portal. I test day to day on V19–V21, and a few conveniences, block renaming among them, need V21. On the older portals expect the core to work and the edges to be thinner.
7 UI languages · EN/DE/IT/ES/FR/RU/ZH · No tokens to buy. Your own key or a local model is optional, on every plan · LAD built through Openness, checked by dry-run validation
Three steps to get started
Download, install and sign in on your engineering PC. Coilmind sets up Openness access itself: one Windows prompt, no manual whitelisting.
Open your project in TIA Portal, then attach Coilmind to it. Openness can only attach to a project that is already open, so TIA stays in charge.
Ask in the chat: analyse, generate, clone, translate, test.
If it gets a block or screen wrong, one click puts it back.
You run production lines, and a wrong change is downtime. An assistant that acts on its own needs a way back, and this is how it works.
For freelancers and small integrators
Whether you freelance or work in a three-person integrator, the same jobs come round every week: cloning blocks, working out someone else’s logic, translating for the client’s HMI, fixing legacy code on site with the clock running.
Coilmind takes the repetitive part of the build, so one person can take on more projects and ship sooner.
Shared templates keep everyone’s code to the same standard, so a colleague can pick up a block someone else generated.
reverse-engineer a handover · clone for a new line · translate for the client · fix legacy on site, offline
Built by one automation engineer
I spent years writing PLC programs for Siemens hardware, and a lot of those years went on the same jobs done by hand: cloning blocks for the twelfth conveyor, reverse-engineering someone else’s STL, retyping tag names. Coilmind started as a tool I built for myself.
Coilmind is not a funded startup looking for an exit. svghmi.pro has been selling tools to Siemens engineers, and it runs on subscriptions, with no funding rounds behind it. So Coilmind gets built in the direction users ask for. There is nobody else I answer to.
If something breaks, you email me and I answer it myself. There is no support team in between.
support@svghmi.pro
Three plans, priced per licence
Every plan is the full product, with AI included and no tokens to buy. The plans differ in volume. Upgrade, downgrade or cancel any time from your account page.
A task is one thing you ask for: "clone this FB for the second drive", "rewrite this STL block in SCL". Coilmind may make a dozen internal calls to finish it, and you’re billed for one task. Fair use is metered over a rolling 7-day window, so the monthly figure is an average rather than a burst allowance.
Check the details yourself
Common questions
Which TIA versions and OS?+
Windows only, because the Openness API is Windows-bound. TIA Portal V17 through V21, including the older versions that newer cloud assistants don’t support. I test day to day on V19–V21 and a few conveniences need V21, so on V17 and V18 expect the core to work and the edges to be thinner.
One thing to know before you download: Openness can only attach to a project that is already open in TIA Portal. Coilmind cannot open a project for you. You open it in TIA, then attach.
Can an LLM really write LAD, or is that just marketing?+
LAD is graphical and a language model only emits text, so pasting pseudo-ladder and hoping was never going to work. Coilmind generates the logic, builds the block itself through the Openness API and gates every result on compilation. What lands in your project is valid LAD (or SCL), checked by the same compiler you’d use by hand. Reading and writing LAD reliably is where most of my development time went.
How is this different from the Siemens Eigen Engineering Agent?+
Different products for different buyers. Eigen is £1,548.16 per seat per year, billed annually with no monthly option, and it needs TIA Portal V19 or newer plus an Industry Organization Account, which is a company-level thing. It does a lot Coilmind doesn’t: hardware, SINAMICS drives, PROFINET, creating WinCC Unified HMI devices.
Coilmind is €29 to €149 a month, or yearly if you prefer, bought by one person with a card, and it works from V17. If you’re an organisation inside the Siemens ecosystem, theirs fits your procurement better. If you buy your own tools, that path is mostly closed to you.
→ Full comparisonIsn’t this the same as the TIA Portal MCP servers people are building?+
The API underneath is similar. Those projects are middleware: you run a local server, drive it from a separate tool such as VS Code, Cursor or Claude Desktop, bring your own API key and pay the model provider yourself.
Coilmind is one application with AI included: nothing else to install, no key to manage. The trade-off is that middleware gives you more raw API surface for building your own workflows. Coilmind is a finished tool.
Can two of us use one licence?+
Yes. A licence covers several machines (two on Starter, four on Basic, ten on Premium) and you can work at the same time. The ceiling is requests per minute; there are no named seats. If you keep hitting the rate limit, that’s the point to move up a plan.
Can it work offline / air-gapped?+
Yes, within a window. After activation a local model runs entirely on your machine and your code never leaves the PC. The licence has a 14-day offline grace period: work as long as you like with no connection, and within those 14 days reconnect briefly so it can refresh. That covers a commissioning trip or a stint on an isolated network. It is not permanent air-gapped operation, and no part of the product is: the licence check comes due either way.
Two more caveats. Local models are generally weaker than the cloud, and how capable they are depends on your GPU budget. The block templater and the generator need no model at all, online or offline.
Do I need my own API key?+
No. AI is included. The provider keys live on our server, you pay one subscription, and there are no tokens to buy and nothing to configure on the model side.
Your own key or a local model is optional, on every plan.
Can I use it without AI at all?+
Yes. The Templater is fully local and deterministic, with no model calls, which helps where the security policy forbids AI.
Does my project content get sent to a model?+
With cloud AI, your prompt and the project context it needs go to our inference providers, DeepInfra and Fireworks, and the answer comes straight back. Both are in the United States, so for an EU customer this is a transfer to the US; our privacy policy names them and our DPA sets out the basis. We keep no copy of your prompts or your project on our side.
Neither provider uses the traffic to train models without an explicit opt-in, and we have not given one. The caveat to know before your client’s IT asks: DeepInfra states it may hold API inputs and outputs briefly for debugging. Nobody keeps them long-term.
If you don’t want any of that, use offline mode. With a local model nothing leaves the PC at all. The templater and the generator involve no model in the first place.
Will AI break my blocks?+
Four things stand between the model and your project. Generated logic can be validated without being written at all: topology, schema and semantics are checked as a dry run. A local snapshot is taken before every block and HMI screen change, and restore is one click. A compile gate runs on every block import: if it doesn’t build, the change isn’t applied. And everything the assistant did is listed in a changelog inside the project.
One gap: snapshots cover blocks and screens, not tag, tag-table or UDT operations. Those you undo in TIA the usual way.
Why is this not just another AI wrapper?+
It talks to TIA through the Openness API instead of working on uploaded files or pasted code, so it sees the project’s actual cross-references, UDTs and tag tables. If that difference doesn’t matter for your work, Coilmind isn’t for you.
What languages?+
The UI ships in 7 languages (EN/DE/IT/ES/FR/RU/ZH). The AI answers in the language you ask in, and translation keeps the PLC context, so no machine pidgin.
How does pricing and the trial work?+
Three plans (Starter, Basic, Premium), paid monthly or yearly by card. No annual lock-in, no org account, AI included. Every plan starts with a 14-day free trial with 25 tasks included and no credit card. You add a card only if you decide to keep it. Checkout is handled by PayPro Global (VAT included), and you can upgrade, downgrade or cancel yourself from your account page.