TL;DR

Pig is an API for launching and interacting with Windows virtual desktops.

Teams building automations can plug Pig into their AI agents, giving them the ability to operate computers as humans would, and perform high-value automations that were previously not possible.

Sign up here

The Problem

Many enterprise tools, from electronic health records to manufacturing supply-chain apps, are desktop-based and do not have a public API, preventing automation.

Traditional hardcoded RPA automations can be brittle. Subtle changes in the desktop apps, like popups or slow loading, can derail an automation. And complex workflows which require planning and judgment are not possible.

As a result, human office workers are employed to do highly repetitive tasks, often involving transferring text between different desktop apps, spreadsheets, and emails.

Our Solution

https://youtu.be/xGli0TG7s3Y

In this example, we show what’s possible building on Pig. We start a virtual desktop on Pig, and hand the connection to an agent loop running the Claude Computer-Use model.

The model then has a high-level task, and it begins to work, using Pig’s human-mimicking APIs such as:

• Screenshotting the desktop screen
• Clicking at coordinates
• Entering keystrokes

All of which are enough to complete the task.

The Pig SDK gives you tools to manage VMs and send interactions to the desktop.

And importantly, Pig includes tools for agents to pass control to human operators during critical operations, such as clicking send on important emails, inputting payment information, or solving captchas.

We intend to be the personal computers for AI, and the embeddable observability window for your end users.

Use Cases

Killer use-cases of Pig involve the following factors:

• the workflow requires a Windows desktop app
• The app does not have a public API

I expect to be surprised by what people build on Pig, but I’m most optimistic for use of:

• Office Suite
• Electronic Health Record Software
• Claims Processing Software
• Factory Control Systems
• Lab Software

Team & Background

I’m Erik, a second-time founder, with five years spent building cloud infrastructure.

Most notably, I built Banana, a serverless GPU product used by more than three thousand teams to host production inference jobs.

Prior to Banana, I built a cloud desktop product for accelerating 3D design apps. Following Banana, I had the privilege to work with our former competitor, Modal, on their serverless Python cloud.

Why am I building Pig?

Throughout 2024, I spent my time experimenting with new technologies and building up a thesis for what the future of AI tooling would be. After poking around a few wild ideas (such as data centers in space and internet-over-laser), I boomeranged back to the area that I know and love: compute infrastructure.

An AI founder friend called me up in Nov 2024, pitched me on this idea as something he needed to see in the world, and said, “Erik, you’re the single best founder to build this.”

And I agree. I am.

How to become a user

Pig just launched our Alpha program, in which I work directly with early users to get their automations live on Pig.

Sign up at pig.dev/alpha and I’ll reach out when Pig is ready for you!

And if you don’t need Pig, please tell at least one friend about it.

Oink!
Erik