Configuration Overview

All operator configuration options: command-line flags and feature toggles

The operator is configured through command-line flags, which map directly to Helm values when deploying via Helm. The core controllers and the self-healing loop are on by default; AI and the optional integrations are opt-in.

​

Feature matrix

Feature	Default	Flag	Description
ApplicationPersona controller	Enabled	Always on	Validates deployments against persona constraints
ClusterPersona controller	Enabled	Always on	Discovers cluster state and resource capacity
Self-healing loop	Enabled	--enable-health-check	Detection, diagnosis, and remediation proposals
AI diagnosis	Disabled	--llm-provider	LLM-enhanced root-cause analysis (requires an API key)
AI remediation planner	Enabled when a provider and key are set	--enable-ai-remediation	Ordered, AI-written remediation plans (Anthropic only)
Metrics-server integration	Enabled	--enable-metrics-server	Container-level CPU/memory signals
ClusterPersona bootstrap	Enabled	--auto-create-cluster-persona	Auto-creates dorgu-cluster on startup
ArgoCD watcher	Enabled	--enable-argocd	Watches ArgoCD Applications for sync status
Prometheus integration	Disabled	--prometheus-url	Queries Prometheus for resource baselines
Event pipeline	Disabled	--enable-event-pipeline	Kubernetes-Event watcher and DorguEvent records. Nothing reads them yet, so it is off since v0.13.0.
Leader election	Disabled (flag default; enabled in Helm chart)	--leader-elect	HA mode for multi-replica deployments
Secure metrics	Enabled	--metrics-secure	HTTPS metrics endpoint with RBAC

​

Command-line flags

​

Core

Flag	Type	Default	Description
--metrics-bind-address	string	0	Metrics endpoint bind address. Use :8443 for HTTPS or :8080 for HTTP. Set to 0 to disable.
--health-probe-bind-address	string	:8081	Health and readiness probe bind address
--leader-elect	bool	false	Enable leader election for HA deployments
--enable-http2	bool	false	Enable HTTP/2 for the metrics server

Operator v0.12.0 removed the Deployment validating webhook and its flags: --enable-webhook, --webhook-mode, --webhook-cert-path, --webhook-cert-name and --webhook-cert-key. Passing one stops the binary from starting, so remove them from any manifest that runs the manager directly. See Helm values for the chart side.

​

Metrics

Flag	Type	Default	Description
--metrics-secure	bool	true	Serve metrics over HTTPS with RBAC
--metrics-cert-path	string	""	Directory containing metrics TLS certificate
--metrics-cert-name	string	tls.crt	Metrics certificate filename
--metrics-cert-key	string	tls.key	Metrics private key filename

​

Self-healing and AI

Flag	Type	Default	Description
--enable-health-check	bool	true	Enable the health-check reconciler: detection, diagnosis, and remediation proposals. When false, the operator records persona health only and never opens an incident or proposes a fix. Defaulted to false in the binary before v0.12.0.
--health-check-interval	duration	60s	Detection cycle interval
--readiness-threshold	duration	2m	How long a running container may stay not Ready before a ReadinessNotReady incident (v0.13.0)
--liveness-restart-count	int	2	Restarts within --liveness-restart-window that raise LivenessRestarts or RestartLoop (v0.13.0)
--liveness-restart-window	duration	10m	The restart window. Must be at least twice --health-check-interval, or the operator refuses to start (v0.13.0)
--incident-recurrence-cap	int	3	Recurrences of one fault within --incident-recurrence-window before the latest incident is reopened instead of a new one created. 0 disables the cap (v0.13.0)
--incident-recurrence-window	duration	24h	The recurrence window (v0.13.0)
--resolved-incident-retention	duration	720h	Resolved incidents older than this are deleted, except the newest of each fault and any a live RemediationAction names. 0 keeps them. Must not be shorter than the recurrence window (v0.13.0)
--remediation-pending-ttl	duration	24h	A Pending proposal older than this becomes Expired. 0 disables the TTL (v0.13.0)
--enable-metrics-server	bool	true	Enable metrics-server integration for container-level detection
--llm-provider	string	""	claude or gemini for AI-enhanced diagnosis. Empty disables AI.
--llm-api-key	string	""	API key, overriding the ANTHROPIC_API_KEY / GEMINI_API_KEY env vars
--llm-model	string	""	Model override for the provider
--llm-cache-ttl	duration	30m	How long an AI diagnosis or plan is reused while the incident’s signals are unchanged. 0 asks the model every cycle (v0.13.0)
--safety-excluded-namespaces	string	""	Comma-separated namespaces Dorgu never proposes a remediation in, on top of kube-system and its own namespace (read from POD_NAMESPACE). See remediation safety.
--enable-ai-remediation	bool	true	Enable AI-written ordered remediation plans. Requires --llm-provider=claude and a key; falls back to the deterministic rules otherwise.

The Helm chart never passes --llm-api-key. It injects the key as an environment variable from a Secret via secretKeyRef (ANTHROPIC_API_KEY for claude), so it stays out of the pod spec. See AI setup.

​

Bootstrap

Flag	Type	Default	Description
--auto-create-cluster-persona	bool	true	Auto-create a ClusterPersona named dorgu-cluster if none exists on startup
--cluster-persona-ensure-interval	duration	2m	How often to re-ensure that persona exists. Clamped to a 30s minimum.

​

Integrations

Flag	Type	Default	Description
--enable-argocd	bool	true	Enable ArgoCD Application watching
--prometheus-url	string	""	Prometheus server URL for baseline learning
--enable-event-pipeline	bool	false	Run the Kubernetes-Event watcher and write DorguEvent records (v0.13.0)
--dorgu-event-retention	duration	24h	How long a DorguEvent is kept. Used only with --enable-event-pipeline
--dorgu-event-max-records	int	2000	Maximum DorguEvent records kept. 0 drops the cap. Used only with --enable-event-pipeline

Operator v0.13.0 removed the WebSocket server and its flags, --enable-websocket and --websocket-addr. Passing either stops the binary from starting with flag provided but not defined. See the upgrade guide.

The ArgoCD watcher only activates if both --enable-argocd=true and the ArgoCD Application CRD (argoproj.io/v1alpha1) is detected in the cluster. If the CRD is missing, the watcher silently skips registration.

​

Example: Full-featured deployment

./bin/manager \
  --metrics-bind-address :8443 \
  --health-probe-bind-address :8081 \
  --leader-elect \
  --enable-health-check \
  --health-check-interval 30s \
  --llm-provider claude \
  --enable-ai-remediation \
  --enable-argocd \
  --prometheus-url http://prometheus-server.monitoring:9090 \
  --safety-excluded-namespaces payments,ingress-nginx

Helm values

Complete Helm values.yaml reference

Installation

Install methods and verification