feat(ollama): add Ollama chart

README.md

@@ -130,6 +130,7 @@ Below is a list of available charts along with their links:
 | **JupyterHub** | [helm.zop.dev/jupyterhub](https://helm.zop.dev/src/readme.html?id=jupyterhub) | <a href="https://zop.dev/zopday/app/deploy?install=jupyterhub"><img src="https://zop.dev/deploytozopday-inkhard.svg" alt="Deploy to Zopday" height="28"></a> |
 | **LiteLLM** | [helm.zop.dev/litellm](https://helm.zop.dev/src/readme.html?id=litellm) | <a href="https://zop.dev/zopday/app/deploy?install=litellm"><img src="https://zop.dev/deploytozopday-inkhard.svg" alt="Deploy to Zopday" height="28"></a> |
 | **LocalAI** | [helm.zop.dev/localai](https://helm.zop.dev/src/readme.html?id=localai) | <a href="https://zop.dev/zopday/app/deploy?install=localai"><img src="https://zop.dev/deploytozopday-inkhard.svg" alt="Deploy to Zopday" height="28"></a> |
+| **Ollama** | [helm.zop.dev/ollama](https://helm.zop.dev/src/readme.html?id=ollama) | <a href="https://zop.dev/zopday/app/deploy?install=ollama"><img src="https://zop.dev/deploytozopday-inkhard.svg" alt="Deploy to Zopday" height="28"></a> |
 | **Outline** | [helm.zop.dev/outline](https://helm.zop.dev/src/readme.html?id=outline) | <a href="https://zop.dev/zopday/app/deploy?install=outline"><img src="https://zop.dev/deploytozopday-inkhard.svg" alt="Deploy to Zopday" height="28"></a> |
 | **Superset** | [helm.zop.dev/superset](https://helm.zop.dev/src/readme.html?id=superset) | <a href="https://zop.dev/zopday/app/deploy?install=superset"><img src="https://zop.dev/deploytozopday-inkhard.svg" alt="Deploy to Zopday" height="28"></a> |
 | **WordPress** | [helm.zop.dev/wordpress](https://helm.zop.dev/src/readme.html?id=wordpress) | <a href="https://zop.dev/zopday/app/deploy?install=wordpress"><img src="https://zop.dev/deploytozopday-inkhard.svg" alt="Deploy to Zopday" height="28"></a> |

charts/ollama/Chart.yaml

@@ -0,0 +1,12 @@
+apiVersion: v2
+appVersion: "0.32.9"
+description: Helm chart for deploying Ollama, a self-hosted LLM inference server
+name: ollama
+version: 0.0.1
+type: application
+icon: "https://storage.googleapis.com/zopdev-blog-resources/1/files/originals/20260818/c8e9597d-0ced-4880-9335-f4f8a78f9f31-ollama.png"
+maintainers:
+  - name: ZopDev
+    url: zop.dev
+annotations:
+  type: application

charts/ollama/README.md

@@ -0,0 +1,211 @@
+# Ollama Helm Chart
+
+[Ollama](https://github.com/ollama/ollama) runs LLMs (Llama 3, Mistral,
+etc.) locally and serves them over a REST API. This chart deploys a
+single Ollama server backed by a persistent volume for pulled models.
+
+---
+
+## Prerequisites
+
+- Kubernetes 1.19+
+- Helm 3+
+
+---
+
+## Add Helm Repository
+
+```bash
+helm repo add zopdev https://helm.zop.dev
+helm repo update
+```
+
+---
+
+## Install Helm Chart
+
+```bash
+helm install my-ollama zopdev/ollama
+```
+
+To pull one or more models automatically on first start:
+
+```bash
+helm install my-ollama zopdev/ollama \
+  --set models={llama3,mistral}
+```
+
+Without `models` set, the server comes up with no models installed —
+pull one afterwards with:
+
+```bash
+kubectl exec my-ollama-ollama-0 -- ollama pull llama3
+```
+
+---
+
+## Uninstall Helm Chart
+
+```bash
+helm uninstall my-ollama
+```
+
+The PersistentVolumeClaim created from the volume template outlives the
+release. Delete it separately to reclaim the disk:
+
+```bash
+kubectl delete pvc models-my-ollama-ollama-0
+```
+
+---
+
+## Configuration
+
+| **Input** | **Type** | **Description** | **Default** |
+|---|---|---|---|
+| `image.repository` | `string` | Container image. | `ollama/ollama` |
+| `image.tag` | `string` | Image tag. | `0.32.9` |
+| `image.pullPolicy` | `string` | Image pull policy. | `IfNotPresent` |
+| `service.type` | `string` | Service type for the API. | `ClusterIP` |
+| `service.port` | `int` | Service port for the API. | `11434` |
+| `diskSize` | `string` | Size of the models volume. | `"20Gi"` |
+| `resources` | `object` | CPU and memory for the pod. | 500m/2Gi – 2000m/4Gi |
+| `models` | `list` | Models to pull automatically on first start. See *Model pulling* below. | `[]` |
+| `nodeSelector` | `object` | Node labels to constrain scheduling. | `{}` |
+| `tolerations` | `list` | Tolerations to allow scheduling onto tainted nodes. | `[]` |
+| `env` | `object` | Extra environment variables. | `{}` |
+| `ingress.enabled` | `bool` | Create an Ingress for the API. | `false` |
+| `ingress.className` | `string` | IngressClass name. | `""` |
+| `ingress.host` | `string` | Hostname. Required when the ingress is enabled. | `""` |
+| `ingress.annotations` | `object` | Ingress annotations. | `{}` |
+| `ingress.tlsSecretName` | `string` | Existing TLS secret for the host. | `""` |
+
+### Model pulling
+
+`models` is a plain list of model names, e.g. `["llama3", "mistral"]`.
+When it's non-empty, an init container starts a throwaway Ollama server
+against the same volume the main container uses, pulls each model in
+order, then exits — the main container only starts once every model is
+already on disk. The init container fails (rather than reporting
+success) if the server doesn't come up within 60s, or if any model
+name is invalid or fails to pull — so a typo in `models` shows up as a
+failed release instead of a healthy-looking pod silently serving no
+model. Pulling several large models can still take a while; only the
+server's own startup is time-bounded, not the pull itself, so a slow
+but working pull is not killed.
+
+Changing `models` on an existing release and running `helm upgrade`
+pulls any newly-added models but does not remove ones taken off the
+list — they stay on the volume until deleted manually.
+
+### Running on GPU nodes
+
+This chart adds no Ollama-specific GPU field. Target a GPU node pool the
+same way as any other workload, through the generic `nodeSelector` and
+`tolerations` passthrough, and request the GPU itself through
+`resources.limits`:
+
+```bash
+helm install my-ollama zopdev/ollama \
+  --set resources.limits."nvidia\.com/gpu"=1 \
+  --set nodeSelector.cloud\\.google\\.com/gke-accelerator=nvidia-tesla-t4
+```
+
+Requires the appropriate device plugin (e.g. NVIDIA's) already installed
+on the cluster.
+
+### Security
+
+Ollama has no built-in authentication — upstream ships no auth
+mechanism at all, so `GET /api/tags` and every other endpoint answer
+any caller with no credentials. That's fine reachable only inside the
+cluster, but turning on `ingress.enabled` publishes an inference API
+(and `POST /api/pull`, which downloads arbitrary models onto the
+volume) to anyone who can resolve the host. Pair an ingress with
+authentication at the proxy/gateway layer, or restrict it to trusted
+networks.
+
+The container also runs as root, matching the upstream image's own
+default — `/root/.ollama` is where it expects to write, so this is a
+deliberate fit rather than an oversight.
+
+---
+
+## Example `values.yaml`
+
+```yaml
+models:
+  - llama3
+  - mistral
+
+resources:
+  requests:
+    cpu: "2000m"
+    memory: "8Gi"
+  limits:
+    cpu: "4000m"
+    memory: "16Gi"
+
+diskSize: "50Gi"
+
+ingress:
+  enabled: true
+  className: nginx
+  host: ollama.example.com
+```
+
+```bash
+helm install my-ollama zopdev/ollama -f values.yaml
+```
+
+---
+
+## Features
+
+- Persistent model storage, so pulled models survive pod restarts
+- Optional automatic model pulling on first start via `models`
+- Generic `nodeSelector`/`tolerations` passthrough for targeting GPU nodes
+- Ingress with optional TLS
+- Configurations that cannot work (a bad ingress setup) are rejected when
+  the chart renders, naming the cause, rather than installing and
+  crash-looping
+
+---
+
+## Connection Config
+
+The API is served on `service.port` (11434) of the `<release>-ollama`
+service.
+
+```bash
+kubectl port-forward svc/my-ollama-ollama 11434:11434
+curl http://localhost:11434/api/generate \
+  -d '{"model": "llama3", "prompt": "hello"}'
+```
+
+- **`/api/generate`**, **`/api/chat`**, **`/api/embeddings`** — inference endpoints.
+- **`/api/pull`** — pull a model over the API instead of `kubectl exec`.
+- **`/api/tags`** — list models currently pulled.
+- **`/`** — used for readiness/liveness; returns `Ollama is running`.
+
+---
+
+## Contributing
+
+We welcome contributions to improve this Helm chart. Please refer to the
+[CONTRIBUTING.md](../../CONTRIBUTING.md) file for contribution
+guidelines.
+
+---
+
+## Code of Conduct
+
+To maintain a healthy and collaborative community, please adhere to our
+[Code of Conduct](../../CODE_OF_CONDUCT.md).
+
+---
+
+## License
+
+This project is licensed under the [LICENSE](../../LICENSE). Please
+review it for terms of use.

charts/ollama/templates/NOTES.txt

@@ -0,0 +1,27 @@
+Ollama is installed as release {{ .Release.Name }}.
+
+Reach the API:
+
+  kubectl port-forward -n {{ .Release.Namespace }} \
+    svc/{{ include "ollama.fullname" . }} 11434:{{ .Values.service.port }}
+
+  curl http://localhost:11434/api/generate \
+    -d '{"model": "<model>", "prompt": "hello"}'
+{{- if .Values.ingress.enabled }}
+
+Or over the ingress at http{{ if .Values.ingress.tlsSecretName }}s{{ end }}://{{ .Values.ingress.host }}
+{{- end }}
+
+{{- if .Values.models }}
+
+These models were pulled automatically on first start:
+{{- range .Values.models }}
+  - {{ . }}
+{{- end }}
+{{- else }}
+
+No models are installed by default. Pull one after install:
+
+  kubectl exec -n {{ .Release.Namespace }} \
+    {{ include "ollama.fullname" . }}-0 -- ollama pull llama3
+{{- end }}

charts/ollama/templates/_helpers.tpl

@@ -0,0 +1,38 @@
+{{/*
+Expand the name of the chart.
+*/}}
+{{- define "ollama.name" -}}
+{{- default .Chart.Name .Values.nameOverride | trunc 63 | trimSuffix "-" }}
+{{- end }}
+
+{{/*
+Fully qualified app name, used for the Service and workload names.
+*/}}
+{{- define "ollama.fullname" -}}
+{{- if .Values.fullnameOverride }}
+{{- .Values.fullnameOverride | trunc 63 | trimSuffix "-" }}
+{{- else }}
+{{- printf "%s-ollama" .Release.Name | trunc 63 | trimSuffix "-" }}
+{{- end }}
+{{- end }}
+
+{{- define "ollama.chart" -}}
+{{- printf "%s-%s" .Chart.Name .Chart.Version | replace "+" "_" | trunc 63 | trimSuffix "-" }}
+{{- end }}
+
+{{/*
+Common labels.
+*/}}
+{{- define "ollama.labels" -}}
+helm.sh/chart: {{ include "ollama.chart" . }}
+{{ include "ollama.selectorLabels" . }}
+{{- if .Chart.AppVersion }}
+app.kubernetes.io/version: {{ .Chart.AppVersion | quote }}
+{{- end }}
+app.kubernetes.io/managed-by: {{ .Release.Service }}
+{{- end }}
+
+{{- define "ollama.selectorLabels" -}}
+app.kubernetes.io/name: {{ include "ollama.name" . }}
+app.kubernetes.io/instance: {{ .Release.Name }}
+{{- end }}

charts/ollama/templates/headless-service.yaml

@@ -0,0 +1,23 @@
+{{/*
+Governing service for the StatefulSet. A StatefulSet's serviceName is
+meant to name a headless service -- that is what gives the pod a stable
+DNS record. The field is immutable once the object exists, so it is far
+cheaper to get right before release than after.
+*/}}
+apiVersion: v1
+kind: Service
+metadata:
+  name: {{ include "ollama.fullname" . }}-headless
+  labels:
+    {{- include "ollama.labels" . | nindent 4 }}
+spec:
+  type: ClusterIP
+  clusterIP: None
+  publishNotReadyAddresses: true
+  ports:
+    - port: {{ .Values.service.port }}
+      targetPort: http
+      protocol: TCP
+      name: http
+  selector:
+    {{- include "ollama.selectorLabels" . | nindent 4 }}

charts/ollama/templates/ingress.yaml

@@ -0,0 +1,36 @@
+{{- if .Values.ingress.enabled }}
+{{- if not .Values.ingress.host }}
+{{- fail "ingress.enabled requires ingress.host" }}
+{{- end }}
+apiVersion: networking.k8s.io/v1
+kind: Ingress
+metadata:
+  name: {{ include "ollama.fullname" . }}
+  labels:
+    {{- include "ollama.labels" . | nindent 4 }}
+  {{- with .Values.ingress.annotations }}
+  annotations:
+    {{- toYaml . | nindent 4 }}
+  {{- end }}
+spec:
+  {{- if .Values.ingress.className }}
+  ingressClassName: {{ .Values.ingress.className }}
+  {{- end }}
+  {{- if .Values.ingress.tlsSecretName }}
+  tls:
+    - hosts:
+        - {{ .Values.ingress.host | quote }}
+      secretName: {{ .Values.ingress.tlsSecretName }}
+  {{- end }}
+  rules:
+    - host: {{ .Values.ingress.host | quote }}
+      http:
+        paths:
+          - path: /
+            pathType: Prefix
+            backend:
+              service:
+                name: {{ include "ollama.fullname" . }}
+                port:
+                  number: {{ .Values.service.port }}
+{{- end }}

charts/ollama/templates/service.yaml

@@ -0,0 +1,15 @@
+apiVersion: v1
+kind: Service
+metadata:
+  name: {{ include "ollama.fullname" . }}
+  labels:
+    {{- include "ollama.labels" . | nindent 4 }}
+spec:
+  type: {{ .Values.service.type }}
+  ports:
+    - port: {{ .Values.service.port }}
+      targetPort: http
+      protocol: TCP
+      name: http
+  selector:
+    {{- include "ollama.selectorLabels" . | nindent 4 }}