Kentinel

🤖

AI cluster review

An agent reviews the cluster every few minutes and shows a healthy / warning / critical verdict with concrete findings and recommendations — history persisted, trends on the dashboard.

💬

AI assistant

Ask "why is payments-api failing? check its logs" — the agent inspects resources, logs, events, and metrics with read-only tools and streams a grounded answer.

✅

Approval-gated fixes

In assisted mode the agent proposes a change as a reviewable diff, right in the chat. You approve, the server applies. Never autonomous — enforced by RBAC.

🔒

Local-first LLMs

Bundled Ollama by default — no API keys, no cluster data leaves your infrastructure. Or switch to Claude, OpenAI, DeepSeek, or Gemini from the UI.

🔔

Notifications

Discord, Slack, and Teams alerts on status transitions (and recoveries). Transition-based, never spammy, with a configurable severity threshold.

🧰

Full console

Resource browser, Monaco YAML editing, log tailing, in-browser pod terminal, events, Prometheus-backed metrics — dark mode included.

See it

Dashboard with the AI cluster review flagging incidents — The AI review flags a typo’d image tag, an unschedulable pod, and a crash loop — each with a concrete recommendation (staged with make demo-incident).

AI assistant chat — Ask anything about the cluster — the assistant investigates with read-only tools and streams a grounded answer.

AI review history — Every review is persisted: filter by status, see when an incident started and when it recovered.

Pod browser — A full resource browser underneath: pods, deployments, YAML editing, logs, terminal, events.

Cluster events with namespace and type filters.

Settings page — Switch between five LLM providers, wire up Discord/Slack/Teams alerts and the daily report — all from the UI, persisted encrypted.

And the assisted-mode loop end to end — the assistant diagnoses the broken checkout pod, proposes the fix as a reviewable diff, and after an inline human approval the server applies it and verifies the rollout:

Diagnose, propose, approve, fixed — the assisted-mode remediation flow

Quickstart

Runs local-first out of the box — a bundled Ollama and a bundled minimal Prometheus, no API keys required:

helm install kentinel oci://ghcr.io/emreoztoprak/charts/kentinel \
  -n kentinel --create-namespace

kubectl -n kentinel port-forward svc/kentinel-server 8080:80
# open http://localhost:8080

Prefer Docker on your laptop? See Deployment — a single docker compose up works against any cluster your kubeconfig can reach.

By default Kentinel installs in read-only mode: it observes and answers questions but cannot change any resource. Deploy with --set mode=assisted to enable approval-gated remediation, manifest editing, and the pod terminal — details in Security.

Try it: break a shop, watch the AI find it

A demo microservice stack ships with the repo — deploy it healthy, then break it in four realistic ways:

make demo-up        # healthy "shop" namespace: storefront, orders-api, cache
make demo-incident  # four distinct incidents appear
# watch the dashboard flag them, then ask the assistant:
#   "why is payments-api failing? check its logs"
make demo-reset     # back to green (recovery notification!)

KentinelThe AI sentinel for your Kubernetes cluster