Kubeclaw – Scale agents to be your assistant and run K8s
AI agents as K8s primitives with ephemeral RBAC per skill, not in-process monoliths.

Reverse-engineers RBAC from audit logs; solves the 403 cluster-admin doom spiral automatically.
Kubernetes platform engineers, DevSecOps teams, enterprises with RBAC compliance requirements
Kubewarden · Kyverno · Falco
The problem: most clusters run with overly permissive RBAC because getting it right manually is tedious.
You end up with cluster-admin bindings everywhere or spend hours crafting policies by hand.
Audicia flips it — ingest audit logs (file or webhook), normalize the access patterns, and output least-privilege RBAC.
It also diffs observed vs. granted permissions to produce a compliance score.
Everything runs as a controller in your cluster via CRDs. No SaaS, no external dependencies.
GitHub: github.com/felixnotka/audicia Website: audicia.io
Happy to answer any questions about the architecture or approach.
AI agents as K8s primitives with ephemeral RBAC per skill, not in-process monoliths.
Shows attack paths through RBAC graphs when most scanners just flag misconfigurations.
Next.js boilerplate claims LLM-friendly components reduce hallucinations, but ShipFast is cheaper.
Feature-rich Kite dashboard adds AI agent layer, but Kubernetes dashboards are well-served by Lens, Rancher, Kubeflow.
Kernel audit interception builds Landlock profiles iteratively when manual policy writing is painful.
uvx one-liners for 10 log formats with Poisson-paced anomaly scenarios.