(a design partnership, but for nerds only)
We’ve been building orchestration software for 3+ years, and we’re the authors of OpenTaco, a project that helps teams run Terraform securely, collaboratively, and without foot-guns.
A few weeks ago, we were looking at Claude Code and had a “wait… what?” moment.
It’s just a CLI.
Why are we running this locally and ignoring the risks? The agent gets access to run any command on our machines. In effect the LLM can do whatever it wants and read all our files and data.
In the IaC world, running Terraform locally is fine for solo setups, but everyone agrees it breaks down fast for teams. Security, visibility, consistency, all gone. And yet, with coding agents, we’re doing the exact same thing again.
There are a million reasons your security team won’t love you running Claude Code (or any CLI-first agent) locally:
But let’s be honest, as developers, we don’t really care. We’ll YOLO it if it’s the fastest path.
That’s fair.
We’re hardcore YOLOer’s ourseleves, but we think you might want something better.
Catty is a managed terminal environment for coding agents.
Instead of running agents locally, you get: