Helm chart reference¶

The FCC Helm chart lives at charts/fcc/ in the source repository.

Chart: fcc
Chart version: 0.1.0
App version: 1.1.1
Kubernetes: >=1.25.0

This doc is the reference for every configurable knob. For a narrative walkthrough, see kubernetes.md.

Install sources¶

1. Local source tree (dev / contributors)¶

git clone https://github.com/rollingthunderfourtytwo-afk/l2_fcc_agent_team_ext.git
cd l2_fcc_agent_team_ext
helm install fcc ./charts/fcc

2. Sister `fcc-helm` repo (end users, once configured)¶

helm repo add fcc https://rollingthunderfourtytwo-afk.github.io/fcc-helm
helm repo update
helm install fcc fcc/fcc

The sister repo is populated automatically by .github/workflows/helm-mirror.yml on every tag push. See helm-repo-mirror.md for the one-time setup.

3. OCI registry (advanced)¶

helm install fcc oci://ghcr.io/rollingthunderfourtytwo-afk/helm/fcc --version 1.1.1

Values reference¶

The full annotated default is at charts/fcc/values.yaml. Production overlay: values-prod.yaml.

Global¶

Key	Type	Default	Purpose
`global.imageRegistry`	string	`ghcr.io/rollingthunderfourtytwo-afk`	Registry for all 4 images
`global.imageTag`	string	`"1.1.1"`	Tag applied to all 4 images unless overridden per-service
`global.imagePullPolicy`	string	`IfNotPresent`	Pull policy for all services
`global.imagePullSecrets`	list	`[]`	Pull secret names (for private registries)
`global.ai.defaultProvider`	string	`mock`	`mock` / `anthropic` / `openai` / `azure_openai` / `ollama` / `litellm` / `vllm`
`global.ai.anthropicApiKey`	string	`""`	Use `--set-string` or pre-created Secret
`global.ai.openaiApiKey`	string	`""`	Same
`global.ai.azureOpenaiApiKey`	string	`""`	Same
`global.ai.azureOpenaiEndpoint`	string	`""`	Azure endpoint URL
`global.ai.ollamaBaseUrl`	string	`""`	Triggers Ollama auto-detection when set
`global.ai.ollamaDefaultModel`	string	`llama3.2:latest`	Default model for Ollama
`global.ai.litellmDefaultModel`	string	`""`	Triggers LiteLLM auto-detection when set
`global.ai.vllmBaseUrl`	string	`""`	Triggers vLLM auto-detection when set
`global.ai.vllmDefaultModel`	string	`""`	Default model for vLLM

Service account + RBAC¶

Key	Type	Default	Purpose
`serviceAccount.create`	bool	`true`	Create a dedicated ServiceAccount
`serviceAccount.name`	string	`""`	SA name (falls back to `<release>-fcc`)
`serviceAccount.annotations`	object	`{}`	Annotations (useful for AWS IRSA, GCP Workload Identity)
`rbac.create`	bool	`true`	Create Role + RoleBinding (Role-scoped, not ClusterRole)

Backend (WebSocket bridge)¶

Key	Type	Default	Purpose
`backend.enabled`	bool	`true`
`backend.image.repository`	string	`fcc-backend`
`backend.image.tag`	string	`""`	Falls back to `global.imageTag`
`backend.replicas`	int	`1`	Horizontal scale (prod: 2)
`backend.service.port`	int	`8765`	WS + `/health` port
`backend.logLevel`	string	`INFO`	`DEBUG` / `INFO` / `WARNING` / `ERROR`
`backend.resources.requests.cpu`	string	`100m`
`backend.resources.requests.memory`	string	`256Mi`
`backend.resources.limits.cpu`	string	`1000m`
`backend.resources.limits.memory`	string	`1Gi`
`backend.podSecurityContext.runAsNonRoot`	bool	`true`
`backend.podSecurityContext.runAsUser`	int	`1001`
`backend.livenessProbe`	object	TCP :8765	Override full probe spec
`backend.readinessProbe`	object	TCP :8765
`backend.nodeSelector`	object	`{}`
`backend.tolerations`	list	`[]`
`backend.affinity`	object	`{}`

Frontend (React + nginx)¶

Same shape as backend, plus:

Key	Type	Default	Purpose
`frontend.ingress.enabled`	bool	`false`	Enable Ingress resource
`frontend.ingress.className`	string	`""`	e.g. `nginx` / `traefik`
`frontend.ingress.annotations`	object	`{}`
`frontend.ingress.hosts`	list	`[{host: fcc.example.com, ...}]`
`frontend.ingress.tls`	list	`[]`

Streamlit¶

Key	Type	Default	Purpose
`streamlit.enabled`	bool	`true`
`streamlit.app`	string	`persona_explorer.py`	Which app under `apps/streamlit/` to run
`streamlit.service.port`	int	`8501`

Jupyter¶

Key	Type	Default	Purpose
`jupyter.enabled`	bool	`true`
`jupyter.token`	string	`""`	Required in production
`jupyter.service.port`	int	`8888`
`jupyter.persistence.enabled`	bool	`true`	Use a PVC for notebook content
`jupyter.persistence.storageClassName`	string	`""`	Cluster default if empty
`jupyter.persistence.accessMode`	string	`ReadWriteOnce`
`jupyter.persistence.size`	string	`10Gi`
`jupyter.persistence.mountPath`	string	`/app/notebooks`

Common install patterns¶

Local dev (kind / minikube / k3d)¶

helm install fcc ./charts/fcc \
  --set global.imagePullPolicy=Never \
  --set jupyter.persistence.enabled=false

Hosted Anthropic backend + external frontend¶

helm install fcc ./charts/fcc \
  --set global.ai.defaultProvider=anthropic \
  --set-string global.ai.anthropicApiKey="$ANTHROPIC_API_KEY" \
  --set frontend.ingress.enabled=true \
  --set frontend.ingress.className=nginx \
  --set frontend.ingress.hosts[0].host=fcc.example.com

Self-hosted vLLM cluster-internal¶

helm install fcc ./charts/fcc \
  --set global.ai.defaultProvider=vllm \
  --set global.ai.vllmBaseUrl=http://vllm.llm-serving.svc.cluster.local:8000/v1

High-availability backend only (disable other services)¶

helm install fcc ./charts/fcc \
  --set backend.replicas=3 \
  --set frontend.enabled=false \
  --set streamlit.enabled=false \
  --set jupyter.enabled=false

Release lifecycle¶

# Install
helm install fcc ./charts/fcc

# Upgrade
helm upgrade fcc ./charts/fcc --reuse-values --set backend.replicas=3

# Rollback
helm history fcc
helm rollback fcc 1

# Uninstall (PVCs preserved)
helm uninstall fcc

# Uninstall (full wipe, including PVCs)
helm uninstall fcc
kubectl delete pvc -l app.kubernetes.io/instance=fcc