Curie – ship Claude Code agents to Kubernetes with Git push

Curie is an open-source project from curie-eng that ships Claude Code agents onto Kubernetes with a Git push. The project is public on GitHub at github.com/curie-eng/curie and appeared on Hacker News with six points and one comment at the time of this write-up.

The core mechanic is deployment through the same Git workflow teams already use for app code. Instead of wiring a separate agent host, operators push to a repo and the agent runs on Kubernetes. That puts Claude Code sessions behind the cluster’s scheduling, secrets, networking, and rollout model rather than on a laptop or a one-off VM.

For engineers building with coding agents, this closes a gap between local agent use and production-shaped environments. Agent work that needs cluster access, long-lived credentials, or multi-service context can live next to the services it touches. Git as the control plane also means agent configs can go through the same review and CI path as application changes.

The market is crowded with agent runners, IDE plugins, and cloud coding products, but most stop at the developer machine or a managed sandbox. Curie sits closer to platform and DevOps practice: treat the agent as a workload you deploy, not only a chat session you start. That is a different bet from desktop-first Claude Code use and from fully hosted agent APIs that hide the runtime.

Practical next steps are to inspect the curie-eng/curie repository for how it maps Git events to Kubernetes objects, what it assumes about cluster auth and secrets, and whether agent state and logs fit your existing observability stack. Watch whether the project stays focused on Claude Code or broadens to other agent runtimes, and whether early adopters report clean multi-tenant isolation when more than one team shares a cluster.