Sola RPA reads the screen with computer vision. The interesting question is what your data lives in.

What Sola is, without the spin

Sola is a credible, modern take on RPA. The old generation (UiPath, Automation Anywhere, Blue Prism) made you build a workflow by hand in a designer, wiring up selectors and image anchors step by step. Sola flips that: you do the task once, it watches, and it produces a bot. Large language models read the intent of what you did, computer vision locates the elements on screen, and the result runs without API access across the tools an office actually uses. It is a no-code editor pointed at business users, and it leans hard on getting from a recording to a working bot in minutes.

Every guide that currently ranks for this topic stops there: the recording UX, the no-code editor, the customer logos, the time-to-value pitch. None of them name the thing that actually decides whether a bot survives in production, which is what surface it reads to find a field. That is the whole subject below.

Two ways to read a screen

There are only two places a UI automation tool can look to find the box it needs to type into. It can look at the pixels, the rendered image of the screen, and recognize the field visually. Or it can look at the accessibility tree, the structured list of controls every well-behaved application publishes for screen readers and other assistive tech. Sola is in the first camp. Mediar is in the second.

On a clean modern SaaS app, both approaches land in roughly the same place. The divergence shows up the moment your workflow touches software written before APIs were the default.

The anchor fact: how a Mediar step actually resolves a field

Here is the part you can verify yourself. When Mediar runs a recorded step against a SAP GUI screen, it never asks “which pixels look like a text box?” It asks the accessibility tree for the control whose name is Vendor Code, gets the node back, sets its value, and then re-reads the tree to confirm the state changed before moving on.

You do not have to take this on faith. The engine that reads and drives that tree is open source: the Terminator SDK at github.com/mediar-ai/terminator resolves a control by role and name, not by coordinates. And before you automate anything, you can dump the same tree Mediar will read using Microsoft's Accessibility Insights or the classic Inspect.exe, and see for yourself whether the field you care about is named in the tree or only visible as pixels.

Four places a vision-based bot loses the thread on legacy desktop

None of these are knocks on Sola for the surface it was built for. They are simply the spots where recognizing a field from its rendered image is the harder bet, and where reading the published tree is the safer one.

What this is worth when the legacy layer is the bottleneck

The reason any of this matters is that the expensive, stuck workflows in most enterprises are precisely the ones sitting on no-API desktop software. These are real deployments, with the workflow named and the math shown, not press-release numbers.

For a longer read on the line where browser and pixel automation stops and the desktop tree begins, see why legacy desktop apps with no API are the real moat and the maintenance math in what brittle selectors cost a UiPath program.

When Sola is the right call

If your workflows live in modern web apps with clean layouts, and your team wants a no-code recorder that a business user can own end to end, Sola's vision-plus- recording model is a reasonable choice, and its time-to-value pitch is real on that terrain. The honest dividing line is not which tool is “better” in the abstract. It is whether the systems you are trying to automate publish a usable accessibility tree, and whether your pixels are stable enough to recognize from one quarter to the next.

If your data is stuck in SAP GUI, a mainframe terminal, a banking core like Jack Henry or Fiserv or FIS, or an EHR like Epic or Cerner, the accessibility-tree approach is not a nicety. It is usually the only thing on screen that is reliable.