Built for AI to operate

AI-Assisted Embedded HMI Design Turn a sketch, sentence, photograph or whiteboard into a working HMI.

Design HMIs — without writing code. Describe a whole project and let the assistant build it, or ask for one change and watch it land. Everything it makes is an ordinary project you can keep editing by hand.

Download for Windows Watch the AI build an HMI

Windows 10 / 11 · 64-bit · 594 MB · free alpha for evaluation

Drew charger_icon, placed W_ChargerIcon48 × 48 · Welcome screen · x 6, y 10
The Design workspace of EDT HMI Studio: a coffee machine HMI on the canvas, the widget palette on the left, the selected screen's properties on the right.
The whole run is one undo stepChecked against the editor's rules first

An alpha build, for key stakeholders to evaluate — please do not pass it to downstream customers. Expect rough edges, and send back whatever you find.

AI Assistant

Software an AI can truly operate.

EDT HMI Studio is rapid application development (RAD) software built for an AI to operate. It works the same controls you do — screens, widgets, styles, text and languages, drawn pictures, and the values the HMI exchanges with your controller — driving the editor itself, in front of you, from a whole project down to a single word.

  1. A whole project

    From a description: screens, widgets, styles, languages, the values it exchanges with your controller, and the bindings between them.

  2. Screens from a picture

    A sketch, a concept image or a photo of an HMI that already exists — read back to you first, then built at the display's real size.

  3. Edits in place

    Text, colours, layout, icons, bindings — in the open project, as one step you can take back.

  4. Questions

    Ask about the project in front of you. It answers, and changes nothing.

The AI assistant beside the Welcome screen. The request, typed in Traditional Chinese, asks to change the W_GetStarted label's text from Get Started! to GO!. The assistant replies that it changed the shared text resource, and the canvas now reads GO!. Figure 1. A one-word change, asked in Traditional Chinese. Tap to enlarge.
Edit in place

Ask in your own language. The HMI changes.

The request, translated On the Welcome screen, change the W_GetStarted component's Text from “Get Started!” to “GO!”.

The assistant found the text resource behind that label, changed it, and said what else would follow: the text is shared, so anywhere it's used updates too — here, only this one label.

  • It works out what you meant — this label showed a shared entry from the project's text table, and that is where it made the change.
  • It tells you the knock-on effects before you find them yourself.
  • The conversation shows its cost — tokens and dollars, as it goes.
The assistant has drawn a white line-art charger station icon and placed it at the top left of the Welcome screen's blue sidebar. The new image widget, W_ChargerIcon, is selected, and Properties shows its image source charger_icon, position 6, 10 and size 48 by 48. Figure 2. An icon drawn, filed and placed from one request. Tap to enlarge.
Generate assets

It draws the icon, then puts it where you asked.

The request, translated Make a charging-station icon as an Image asset and put it at the top left of Welcome.

It drew a white line-art charger — body, display window, cable, a lightning bolt — in colours chosen for the blue sidebar, filed it as charger_icon in the icons folder, and placed W_ChargerIcon at x 6, y 10.

  • Ordinary project content — select it and Properties shows everything, exactly as if you'd placed it yourself.
  • A real asset, in the Asset library, ready to reuse on any screen.
  • Nothing hidden — no private format, nothing only the AI can edit.
Why you can let it drive

Fast, and still yours.

The assistant is a user of the product, not a second product. It can only do what you could do at the same controls.

Same controls as you

Every move is an editor action, visible on the canvas while it happens.

One undo step

However much it did, a run comes back with a single Undo.

Checked before applied

Each step passes the editor's own rules. A mistake goes back to the model, not into your project.

Never touches hardware

It makes content. Sending an HMI to a device stays your decision.

Your model, your choice

In the cloud, or on your own machine.

Both are offered only models that can read a picture and call tools — the two things the job takes.

Cloud

OpenRouter

Frontier models through your own OpenRouter key. The key stays on your computer; the page never sees it.

  • Claude Fable 5 recommended, with other vision models to choose from
  • Running cost shown in every conversation
  • Remaining credit checked from Preferences
On this machine

Local AI

Run a model locally with Ollama — no key, no cloud, nothing billed. Your project doesn't leave the computer.

  • Any installed model that can see and call tools
  • Same assistant, same undo, same checks
  • The stronger the model, the better the HMI
Rapid application development

From a sentence to a running HMI.

Let the AI draft the HMI, then finish it by hand in the same editor — or the other way round. Nothing about the project changes depending on who made it.

  1. 1Describe or sketch it
  2. 2The AI builds screens and values
  3. 3Refine by hand or by asking
  4. 4Preview it, then put it on the display
No code in any of them

Design, Logic, Protocol. One HMI.

Taken from this build. Tap a picture to open it full size.

Design

What you see is the HMI.

Drag a widget on, set it, arrange it — the palette on the left, everything about the selection on the right.

The Design workspace: an HMI on the canvas, a widget palette on the left, the selected item's properties on the right.
Logic

Behaviour, wired as a diagram.

A payment window counted down, compared and branched — nine nodes, ten wires, no scripting language.

The Logic workspace: a node graph wiring a timer, arithmetic and a comparison to values on the screen.
Protocol

What it exchanges with your controller.

A table of named values — where each lives, how it's read, how often — and a budget that says whether it fits on the wire.

The Protocol workspace: a table of named values with addresses, data types, access and poll intervals.
Watch it work

An AI builds a coffee machine HMI.

A multilingual coffee machine HMI, generated by the assistant and running against a controller — once over serial commands, once over Modbus.

AI-generated HMI over UART serial commands

A multilingual coffee project, generated by the assistant, talking to its controller in serial commands.

Recorded on 0.8.1Serial commandsMultilingual

AI-generated HMI over Modbus

A multilingual coffee machine, generated by the assistant, reading and writing its controller's values over Modbus.

ModbusMultilingual

More walkthroughs

A washing-machine HMI, built and run in the Studio

Recorded on 0.8.4.dev.

The same HMI, earlier

Recorded on 0.8.1.dev.

Download

Try the alpha.

Self-contained — nothing to install first. Bring an OpenRouter key or a local model to wake the assistant; everything else works without one.

Download for Windows SHA-256 checksum 594 MB · Windows 10/11 64-bit

Verify before running:

1c5ea2e29ca1f1fc7c98495aaa3f003de255473efd22bd23fab1b100dea43707 
certutil -hashfile EdtHmiStudio-0.8.5-win-x64-setup.exe SHA256

Not code-signed yet, so SmartScreen will warn. Check the hash, then More info → Run anyway. It's a desktop download — on a phone, send yourself this page and open it on your PC.

What comes next

The road to release.

  1. Now · September 2026

    0.8.5 Alpha

    This build. For evaluation by key stakeholders.

  2. October 2026

    Beta

    Feature-complete and stabilising, with this build's feedback folded in.

  3. December 2026

    General release

    The first build meant for customers — possibly earlier.

Version
0.8.5 Alpha
Released
10 September 2026
Platform
Windows 10 / 11, 64-bit
Installer
594 MB, self-contained
Audience
Key stakeholders only