K8s Sentinel is a Kubernetes security console that runs inside the cluster. It puts Cilium Tetragon runtime monitoring, Cilium network policy enforcement and ValidatingAdmissionPolicy admission control behind one web UI — manage policies, watch security events, route alerts and see live network topology without hand-writing YAML.
Each layer does what only it can do:
| Layer | Responsibility |
|---|---|
| Tracing Policy (Tetragon) | Process execution and file access, with full process context — it can tell a legitimate server process apart from a cryptominer inside the same pod, and kill it |
| Network Policy (Cilium) | Network access control by workload identity — cannot be bypassed by connecting straight to a pod IP; drops packets instead of killing processes |
| Admission Policy (VAP) | Rejects non-compliant resources before they are admitted to the cluster |
- Tracing Policy — process and file whitelist/blacklist rules through a form builder (file rules with read/write scoping and per-process exceptions), with Monitoring / Protect modes and a global switch
- Network Policy — CiliumNetworkPolicy builder with ingress and egress in one policy, whitelist/blacklist per direction, cross-namespace peers and L7 HTTP rules
- Admission Policy — a CEL builder covering nine rule types (labels, annotations, images, replicas, resource limits, security context, host access, ConfigMap and Secret size), plus bindings
- Quarantine — cut a suspect pod off from the network without killing it, so the evidence survives; one click from a Security Event, one to release
- Behavior Discovery — learns what each workload actually executes, and turns it into a policy prefill
- Network Topology — live connection graph from Hubble flows: policy denials in red with the denying policy named, exposure paths traced hop by hop, quarantined pods marked
- Security & Admission Events — persisted, deduplicated, filterable, exportable; webhook alerts (Slack / Teams / Discord) and syslog forwarding fire once per event, paced and retried so a burst does not trip provider rate limits
- Audit Log — every administrative action recorded (who did what, when), filterable, with CSV export
- Event Sources — live ingestion health per source, so a Tetragon agent that is Ready but whose gRPC stream is not actually delivering shows as down rather than falsely healthy
- Dashboard & Settings — agent and ingestion health, event counts, quarantine status; local accounts with Admin / Viewer roles (forced first-login password change, per-IP login rate limiting), retention tuning
- Hardened by default — same-origin Content-Security-Policy and security headers on every response; sessions can survive restarts without persistent storage by injecting
JWT_SECRETfrom a Kubernetes Secret
Requires Kubernetes 1.32+, Cilium (kube-proxy replacement + Hubble Relay), and Tetragon with its gRPC server on the pod network — setup details in docs/install.md.
git clone https://github.com/cooloo9871/K8s_Sentinel.git && cd K8s_Sentinel
# Installs Tetragon and Sentinel from inside the cluster
kubectl apply -f deploy/install-job.yaml
kubectl port-forward -n sentinel-system svc/sentinel 8080:80
# open http://localhost:8080 — admin / admin, change it on first loginImage: ghcr.io/cooloo9871/sentinel — pin a release tag in production.
Technical documentation site: sentineldoc.github.io
In-repo references:
| Doc | Covers |
|---|---|
| docs/install.md | Cilium and Tetragon setup with the flags explained, install options, persistent storage, environment variables, resources, RBAC |
| docs/audit-webhook.md | Wiring the kube-apiserver audit log to Sentinel for full admission-event coverage |
| docs/features.md | Every feature in detail, and what to know before switching on enforcement |
| docs/architecture.md | How the pieces fit together — ingestion pipelines, topology decisions, known limitations |