Status: Proposal
kmx gives developers one command-line entry point for creating and running
Kubernetes agents. It packages the working Kaimahi developer journey without
replacing Kubernetes, kagent, or the agent YAML.
Install the CLI:
brew install kaimahi-agents/tap/kmxUse an existing Kubernetes cluster:
kmx ctx kind-my-clusterOr create a local kind cluster with kagent, Ollama, and Kaimahi:
kmx upCreate and run an agent:
kmx agent create hello-world
kmx agent chat hello-world "What can you do?"kmx agent create uses safe local defaults, writes
agents/hello-world.yaml, applies it to the active development context, and
waits until the agent is ready. The generated YAML remains the source of
truth: developers can edit, diff, review, and reapply it.
| Command | Purpose |
|---|---|
kmx ctx <context> |
Select and validate an existing Kubernetes context |
kmx up |
Create a local kind environment and install its dependencies |
kmx agent create <name> |
Scaffold, apply, and wait for a declarative agent |
kmx agent chat <name> <message> |
Talk to an agent |
kmx status |
Show agents, models, tools, and platform health |
kmx govern <name> |
Enable the Kaimahi governance path for an agent |
kmx down |
Delete the local environment created by kmx up |
kmx is a DevX layer over kind, kubectl, Helm, and kagent. It should:
- make the first successful agent take minutes and a few commands;
- work with either an existing cluster or a disposable local cluster;
- generate portable, reviewable agent YAML rather than hide it;
- show the active context before every mutation and confirm non-local changes;
- delegate runtime-specific operations to kagent or a runtime.
The first milestone is the quick-start path above. Additional Kaimahi
capabilities should become kmx commands only after that path is simple,
reliable, and useful on its own.