Humanity's first attempt to help DevOps engineers with operations.
Original Kagent project: https://kagent.dev/
With this project I explore the practical potential of Kagent in my day-to-day DevOps life. It serves as a testbed for evaluating how AI-powered agents can enhance and streamline real platform engineering tasks.
This setup uses Ollama model; to use other models, refer to the Kagent project.
export KAGENT_DEFAULT_MODEL_PROVIDER=ollama
bash hack/setup.shThe setup script handles all the heavy lifting:
- ✓ Verifies required binaries (k3d, kubectl..)
- ✓ Creates a k3d Kubernetes cluster
- ✓ Installs the Kagent agent and kubectl plugin
- ✓ Deploys Kagent into the cluster
- ✓ Applies all necessary manifests
If you are running local LLM providers (such as Ollama) on a personal laptop or a resource-constrained environment, you may experience heavy memory pressure or context window truncation during multi-step agent tool execution.
To ensure smooth operation and prevent out-of-memory (OOM) crashes, you can constrain the context window size in your model configuration (manifests/01-model-config.yaml):
options:
num_ctx: "2048"