How Creo built an AI Developer for Internal Tools


Felicia Chang

Creo (YC W24) is the quickest way to build internal tools. Their AI agent builds simple one-pager CRUD apps, all in the browser.

The Challenge

The Creo team had learned from hundreds of customer conversations that users wanted an editor that worked in the browser. In this editor, the AI agent could write and compile code in response to natural language prompts.

But in order for a user or AI to write and run sandboxed code, the Creo team needed a way to create isolated processes for every user. This allows the code to run securely and consistently, avoiding any dependencies or compatability issues across devices.

Every agent that writes code or does something meaningful needs to operate in an isolated container and needs to be really user friendly. A user should be able to go into a web app, click a button and the AI should be able to build the thing.

But building the thing is hard. How can an AI develop without an isolated container?

That’s where Jamsocket came in.
Rohan Mayya

The Solution

The Creo team had considered different isolated container solutions, but ultimately chose Jamsocket for its API design and quick customer support. As the Creo team began working with Jamsocket, we set up a Slack channel to work together on their specific use case.

But it was really the simplicity of the Jamsocket API that made the decision easy.

“The API layer is relatively simple – you spawn and stop a container. But I see a lot of companies do this wrong… There are so many things, so many small technical details that I’ve come to realize only by using worse products that do similar things,” said Rohan.

During development, the team took advantage of the Jamsocket Dev CLI, which was “super convenient to test – just spawn a session backend locally, super smooth.”

Deploying their code to production was just as straightforward. With Jamsocket, you can deploy any Dockerized code.

Creo's Session Backend

By using the Jamsocket API to manage the session backend lifecycle, the Creo team was able to focus on the real business logic of their application.

In their Jamsocket session backend, Creo has three components that work together to create the dev environment for the AI agent:

  • A VSCode server that interacts with the editor in the browser to display code that the AI agent has written
  • A NextJS web app that the AI agent has built with the code displayed in the VSCode editor
  • A FastAPI Python server that receives user prompts and retrieves relevant files that the AI agent should make edits to

The Future

In just a few weeks, the team was able to launch their “biggest update yet,” gaining lots of attention on LinkedIn. The team was able to test the product with customers and identify critical insights on how to continue towards the “perfect form factor.”

Since Creo's launch, we've continued to provide technical insights unique to their use case. As Creo works to define the future of internal tools with AI, our job is to abstract away the infrastructure complexity.

If you’re interested in building AI applications on Jamsocket, you can check out and run a demo that uses an OpenAI Assistant to draw on a multiplayer canvas. Or book a time to chat with us.

Ready to build?

Getting started is simple and quick. Explore documentation, or sign up and start building today.