Mediar, the company: a two-person AI lab in San Francisco shipping enterprise desktop automation on top of its own open-source Rust SDK.

First, four companies share a name

Type "mediar" into a search box and you will get back at least four companies, only one of which is the subject of this page. The biggest source of confusion online is MediaRadar, the ad-intelligence tool, which often shows up first because it has been around for fifteen years and spends real money on content. The two biotech and retail-analytics namesakes are smaller but old enough to have built Wikipedia-shaped surface area.

For the rest of this page, "Mediar" means Mediar.ai, Inc., the AI desktop automation company at mediar.ai.

The origin: it didn't start as RPA

The founders did not set out to build a UiPath competitor. The original product was Screenpipe, an open-source desktop intelligence tool Louis shipped in 2023. Screenpipe ran locally on a Mac or Windows machine, captured everything on the screen, OCR'd it, and exposed the result to LLMs as long-running context. Think of it as a personal memory layer for a knowledge worker.

It worked, and it found an audience of developers, but it never settled on a wedge that could carry a company. The breakthrough came from inbound: by mid-2024, several startups (including a few from the Y Combinator network) were asking the same question, which was "can we use the OS-level capture as the backend for our AI operators?" Instead of giving the capture away as a library, the founders wrapped it in an enterprise product and pointed it at the messiest recurring business problem a desktop agent can solve: workflow automation in legacy Windows apps.

That product is Mediar. The capture and execution layer that grew out of Screenpipe got renamed and re-released as terminator-rs, which is the MIT-licensed Rust SDK the commercial product depends on. Same code anyone can read on GitHub.

The timeline at a glance

The whole arc, from the open-source predecessor to a Windows desktop running unattended at a Fortune 500 manufacturer, fits on five labels.

What is open and what is commercial

The interesting structural thing about the company is that the executor underneath the commercial product is a regular Cargo dependency, not a proprietary blob. If you open the Cargo.lock for Mediar's production desktop monorepo you will see this entry verbatim:

[[package]]
name = "terminator-rs"
version = "0.24.31"
dependencies = [...]

That crate, plus the recorder and computer-use crates that sit alongside it, is what does the actual UI Automation work, the focus restoration cascade, and the element lookup at replay. It is published under MIT at github.com/mediar-ai/terminator. Anyone can vendor it, fork it, or build a competing product on top of it without paying Mediar a cent. That is intentional. The defensibility lives in the layers above.

The split tells you what the founders think the company is actually selling. It is not the act of clicking a button on a Windows desktop. It is the recording-to-replay pipeline, the prompt library that turns raw events into semantic intent, the orchestration that runs thousands of those workflows reliably, and the people who get on the phone when something breaks at 2am.

Both founders write production code. Turn-key program customers get their phone numbers and text or call them directly; there is no support tier system or ticketing layer. That is workable at the current customer count and is one of the reasons the company is selling its first hundred customers itself rather than building a sales org.

Why a pre-seed startup is competing with a $10B incumbent

UiPath has a market cap measured in the tens of billions and thousands of certified developers in its partner network. Mediar has two founders and a Cap Table you could fit on a napkin. The honest answer to why this comparison even makes sense in 2026 is that the substrate changed.

The big RPA vendors were built in an era where the only way to make a Windows desktop scriptable was to record DOM-style selectors and replay them deterministically. That worked, but it produced a stack that takes months to deploy, costs six figures up front, requires certified developers, and breaks every time a vendor ships a UI refresh. The maintenance treadmill on a single SAP screen change is the thing that quietly bleeds RPA programs to death.

Two things changed under that stack. First, Windows UI Automation has been quietly improving for fifteen years and now exposes most line-of-business apps cleanly enough to drive them through accessibility APIs alone. Second, LLMs got reliable enough to convert a recorded session into a structured intent record without a human in the loop. Together those two shifts mean a small team can build a more reliable executor than the incumbents shipped, in less time, at a fraction of the cost.

The market hasn't repriced yet. Most enterprise buyers still budget RPA at the UiPath line item. That gap, between what desktop automation costs to deliver in 2026 and what enterprises still pay for it, is the opening Mediar is running through. It is the reason the company can sell a turn-key program for $10K to a customer that was about to sign a $200K UiPath renewal.

What kind of company this is, in shape

Calling Mediar "a startup" is accurate but loses the texture. Three details that matter if you are evaluating the company as a vendor.

First, it is engineering-heavy. Both founders write production code. There are no AEs, no dedicated CSMs, and the demos are run by the people who built the thing being demoed. If you ask a hard technical question on a sales call, you get a real answer or a real "we haven't solved that yet."

Second, it is intentionally close to revenue. The turn-key program exists because the founders would rather earn $10K from a customer in week one than spend a year burning a seed round on marketing. That posture has consequences: the product is shaped by the workflows the first 30 customers actually need, not by what a PM thinks an enterprise might want.

Third, it is open in a meaningful way. The executor at the bottom of the stack is on GitHub under MIT. You can read the implementation of the four-strategy focus restoration cascade and decide for yourself if it works. You can use it without paying Mediar. Most vendors in this category either keep their core proprietary or cosplay openness through a community edition. Mediar publishes the actual crate the production product ships against.

Frequently asked questions