Cotera Code: Running Claude Code Inside VMs to Make Them Play Nicely with Your Other Services

One of the problems that we, like every single company in our space, are experiencing is the overcomplexity of the number of AI services that we use and the cost problems that come from it. Everyone these days is complaining about their cloud costs, especially organizations that are running large amounts of AI agents. Our engineering team is no different. Every day, thousands of dollars leave our bank account, making their way over to Cursor, Devin, OpenAI, or Anthropic, depending on the coding agent that people want to use. Obviously, that can't stand forever because we don't have unlimited dollars.

So what we did was something pretty clever. We baked Claude Code and its subscription model into our own application and created an agent framework that allows Cotera to use Claude Code via a machine in the cloud through Fly.io's Sprites product.

Enter Sprites

Cotera's newest integration is Sprites. What Sprites does, by and large, is very simple. It spins up a virtual VM that scales up or down based on how much you use it. So basically, if it's being used for four hours, it will cost around 40 cents, according to Sprites. This makes it very useful to spin up or down based on what is going on in the machine itself.

Cotera Code

We duplicated our Cotera dev environment in it, launched it into the world, and then created a Sprites tool with comprehensive documentation that allows us to access it from Cotera. This enables us to do a lot of really interesting things. Not only can we now access Claude Code and its subscription model from any device, including our mobile devices, but we can also use Cotera's other integrations to easily take advantage and automate tasks.

What This Unlocks

For example, one of our employees created an agent that checks the Linear triage every morning, finds low-hanging fruit, and starts using the excess Claude Code spend that we have allocated to solve those problems automatically.

Another agent that one of our customer success team members created takes the transcript from every call we have from a customer, or any Slack message from the day, finds any users that are complaining about our product, and if it's simple enough, it makes a Linear ticket, sets off the Cotera Code agent, runs it to fix the problem, sends it to GitHub, and then messages us on Slack.

It's very simple. We're really excited to be able to share this, and we can actually get people set up on top of this very quickly.