Kubernetes AI agent setup with InsForge and Cursor

Abhishek builds a Kubernetes troubleshooting AI agent with a browser dashboard, a FastAPI backend and kubectl-based evidence collection. The tutorial uses Cursor and five repository prompts to create the project structure, investigation engine, LLM reasoning layer and backend integration.

The demo requires Docker running on the machine, a Kubernetes cluster and an InsForge account. A local kind cluster supplies the test environment. InsForge handles authentication, investigation history and progress updates, while its MCP connection lets Cursor configure backend resources. Docker Compose runs the application locally, with the dashboard at localhost:3000. Cluster selection uses the presenter's kubeconfig.

The investigation collects pod status, logs, events, deployment details and network information, then sends that evidence to a model through OpenRouter. Running the app locally therefore does not make the demonstrated reasoning offline. The dashboard returns a proposed root cause, suggested commands and a confidence score.

The failure tests show mixed results. The app identifies an invalid Nginx image tag, but attributes a deliberately crashing container to a node reboot. The presenter acknowledges that the suggested fix will not resolve that crash. Viewers should treat the displayed confidence percentages as model output rather than measured accuracy.