Open-source AI Security Operations Center: alert fusion, LLM-agent triage, MITRE ATT&CK investigation, and a replayable decision ledger for every agent step. Self-hostable, runs with no API keys, MIT licensed. Ships an MCP server for Claude, Cursor and Continue.
Python agentic-aiai-securityalert-triageblue-teamcybersecuritydetection-engineeringincident-responselanggraphllm-agentsmcpmitre-attackmodel-context-protocolpurple-teamsecurity-operationsself-hostedsiemsoarsocthreat-detectionthreat-intelligence
feat(helm): the chart brings up every CORE service (parity 6.8)
Beenu Arorae5a00decec1f
diff --git a/infra/helm/aisoc/templates/core-platform.yaml b/infra/helm/aisoc/templates/core-platform.yamlnew file mode 100644index 000000000..e5d7feb1e--- /dev/null+++ b/infra/helm/aisoc/templates/core-platform.yaml@@ -0,0 +1,289 @@+{{/*+Depth plan 4.5 / parity 6.8 — the three CORE services that are not ours.++These are deliberately NOT entries in `.Values.services`. Everything in+that map is rendered by one generic Deployment template that defaults the+image tag to `.Chart.AppVersion`, and these are third-party images with+their own release cadences: tagging LiteLLM, Ollama or Qdrant with AiSOC's+version produces a chart that renders perfectly and pulls a tag that does+not exist. Two of them also need storage, which a stateless Deployment+template has nowhere to put.++Each is individually toggleable, because a deployment pointing at a hosted+model provider wants LiteLLM and not Ollama, and one with an external+vector store wants neither Qdrant nor its volume.+*/}}++{{- if .Values.llmGateway.enabled }}+---+apiVersion: v1+kind: ConfigMap+metadata:+ name: {{ include "aisoc.fullname" . }}-litellm-config+ namespace: {{ .Release.Namespace }}+ labels:+ {{- include "aisoc.labels" . | nindent 4 }}+ app.kubernetes.io/component: litellm+data:+ config.yaml: |+{{ .Values.llmGateway.config | indent 4 }}+---+apiVersion: apps/v1+kind: Deployment+metadata:+ name: {{ include "aisoc.fullname" . }}-litellm+ namespace: {{ .Release.Namespace }}+ labels:+ {{- include "aisoc.labels" . | nindent 4 }}+ app.kubernetes.io/component: litellm+spec:+ replicas: {{ .Values.llmGateway.replicaCount }}+ selector:+ matchLabels:+ {{- include "aisoc.selectorLabels" . | nindent 6 }}+ app.kubernetes.io/component: litellm+ template:+ metadata:+ labels:+ {{- include "aisoc.selectorLabels" . | nindent 8 }}+ app.kubernetes.io/component: litellm+ annotations:+ checksum/litellm-config: {{ .Values.llmGateway.config | sha256sum }}+ spec:+ serviceAccountName: {{ include "aisoc.serviceAccountName" . }}+ containers:+ - name: litellm+ image: "{{ .Values.llmGateway.image.repository }}:{{ .Values.llmGateway.image.tag }}"+ imagePullPolicy: {{ .Values.llmGateway.image.pullPolicy }}+ args: ["--config", "/etc/litellm/config.yaml", "--port", "4000"]+ ports:+ - name: http+ containerPort: 4000+ env:+ {{- range $key, $value := .Values.llmGateway.env }}+ - name: {{ $key }}+ value: {{ $value | quote }}+ {{- end }}+ {{/*+ The gateway serves /health/liveliness 200 while failing every+ completion when its request timeout resolves to nothing, so+ liveness alone is not evidence it works -- it is evidence the+ process is up, which is all a liveness probe should claim.+ */}}+ livenessProbe:+ httpGet:+ path: /health/liveliness+ port: http+ initialDelaySeconds: 20+ periodSeconds: 20+ readinessProbe:+ httpGet:+ path: /health/readiness+ port: http+ initialDelaySeconds: 10+ periodSeconds: 10+ resources:+ {{- toYaml .Values.llmGateway.resources | nindent 12 }}+ volumes:+ - name: config+ configMap:+ name: {{ include "aisoc.fullname" . }}-litellm-config+---+apiVersion: v1+kind: Service+metadata:+ name: {{ include "aisoc.fullname" . }}-litellm+ namespace: {{ .Release.Namespace }}+ labels:+ {{- include "aisoc.labels" . | nindent 4 }}+ app.kubernetes.io/component: litellm+spec:+ type: ClusterIP+ ports:+ - port: 4000+ targetPort: http+ name: http+ selector:+ {{- include "aisoc.selectorLabels" . | nindent 4 }}+ app.kubernetes.io/component: litellm+{{- end }}++{{- if .Values.ollama.enabled }}+---+apiVersion: apps/v1+kind: StatefulSet+metadata:+ name: {{ include "aisoc.fullname" . }}-ollama+ namespace: {{ .Release.Namespace }}+ labels:+ {{- include "aisoc.labels" . | nindent 4 }}+ app.kubernetes.io/component: ollama+spec:+ serviceName: {{ include "aisoc.fullname" . }}-ollama+ {{/*+ One replica, and a StatefulSet rather than a Deployment. The model+ weights are several gigabytes on a volume; two replicas behind one+ Service would each pull their own copy and the first request after a+ rollout would block for minutes on a cold one.+ */}}+ replicas: 1+ selector:+ matchLabels:+ {{- include "aisoc.selectorLabels" . | nindent 6 }}+ app.kubernetes.io/component: ollama+ template:+ metadata:+ labels:+ {{- include "aisoc.selectorLabels" . | nindent 8 }}+ app.kubernetes.io/component: ollama+ spec:+ serviceAccountName: {{ include "aisoc.serviceAccountName" . }}+ containers:+ - name: ollama+ image: "{{ .Values.ollama.image.repository }}:{{ .Values.ollama.image.tag }}"+ imagePullPolicy: {{ .Values.ollama.image.pullPolicy }}+ ports:+ - name: http+ containerPort: 11434+ env:+ - name: OLLAMA_HOST+ value: "0.0.0.0"+ {{- range $key, $value := .Values.ollama.env }}+ - name: {{ $key }}+ value: {{ $value | quote }}+ {{- end }}+ {{/*+ A generous initialDelay: the first start pulls the model named+ in `.Values.ollama.model`, which is gigabytes. A probe tuned+ for a stateless service kills the pod mid-pull, forever.+ */}}+ startupProbe:+ httpGet:+ path: /+ port: http+ periodSeconds: 10+ failureThreshold: {{ .Values.ollama.startupFailureThreshold }}+ readinessProbe:+ httpGet:+ path: /+ port: http+ periodSeconds: 15+ resources:+ {{- toYaml .Values.ollama.resources | nindent 12 }}+ volumeMounts:+ - name: models+ mountPath: /root/.ollama+ volumeClaimTemplates:+ - metadata:+ name: models+ spec:+ accessModes: ["ReadWriteOnce"]+ {{- with .Values.ollama.persistence.storageClass }}+ storageClassName: {{ . }}+ {{- end }}+ resources:+ requests:+ storage: {{ .Values.ollama.persistence.size }}+---+apiVersion: v1+kind: Service+metadata:+ name: {{ include "aisoc.fullname" . }}-ollama+ namespace: {{ .Release.Namespace }}+ labels:+ {{- include "aisoc.labels" . | nindent 4 }}+ app.kubernetes.io/component: ollama+spec:+ type: ClusterIP+ ports:+ - port: 11434+ targetPort: http+ name: http+ selector:+ {{- include "aisoc.selectorLabels" . | nindent 4 }}+ app.kubernetes.io/component: ollama+{{- end }}++{{- if .Values.qdrant.enabled }}+---+apiVersion: apps/v1+kind: StatefulSet+metadata:+ name: {{ include "aisoc.fullname" . }}-qdrant+ namespace: {{ .Release.Namespace }}+ labels:+ {{- include "aisoc.labels" . | nindent 4 }}+ app.kubernetes.io/component: qdrant+spec:+ serviceName: {{ include "aisoc.fullname" . }}-qdrant+ replicas: 1+ selector:+ matchLabels:+ {{- include "aisoc.selectorLabels" . | nindent 6 }}+ app.kubernetes.io/component: qdrant+ template:+ metadata:+ labels:+ {{- include "aisoc.selectorLabels" . | nindent 8 }}+ app.kubernetes.io/component: qdrant+ spec:+ serviceAccountName: {{ include "aisoc.serviceAccountName" . }}+ containers:+ - name: qdrant+ image: "{{ .Values.qdrant.image.repository }}:{{ .Values.qdrant.image.tag }}"+ imagePullPolicy: {{ .Values.qdrant.image.pullPolicy }}+ ports:+ - name: http+ containerPort: 6333+ - name: grpc+ containerPort: 6334+ readinessProbe:+ httpGet:+ path: /readyz+ port: http+ initialDelaySeconds: 10+ periodSeconds: 10+ livenessProbe:+ httpGet:+ path: /livez+ port: http+ initialDelaySeconds: 20+ periodSeconds: 20+ resources:+ {{- toYaml .Values.qdrant.resources | nindent 12 }}+ volumeMounts:+ - name: storage+ mountPath: /qdrant/storage+ volumeClaimTemplates:+ - metadata:+ name: storage+ spec:+ accessModes: ["ReadWriteOnce"]+ {{- with .Values.qdrant.persistence.storageClass }}+ storageClassName: {{ . }}+ {{- end }}+ resources:+ requests:+ storage: {{ .Values.qdrant.persistence.size }}+---+apiVersion: v1+kind: Service+metadata:+ name: {{ include "aisoc.fullname" . }}-qdrant+ namespace: {{ .Release.Namespace }}+ labels:+ {{- include "aisoc.labels" . | nindent 4 }}+ app.kubernetes.io/component: qdrant+spec:+ type: ClusterIP+ ports:+ - port: 6333+ targetPort: http+ name: http+ - port: 6334+ targetPort: grpc+ name: grpc+ selector:+ {{- include "aisoc.selectorLabels" . | nindent 4 }}+ app.kubernetes.io/component: qdrant+{{- end }}diff --git a/infra/helm/aisoc/values.yaml b/infra/helm/aisoc/values.yamlindex 7b687804c..b52fa7591 100644--- a/infra/helm/aisoc/values.yaml+++ b/infra/helm/aisoc/values.yaml@@ -250,6 +250,104 @@ services:extraVolumeMounts: []# ─── Ingress ──────────────────────────────────────────────────────────────────+ # Depth plan 4.5 / parity 6.8. These three were absent from the chart+ # while being CORE services in docker-compose.yml, so `helm install`+ # brought up a deployment that was missing response execution, every+ # connector poll and the threat-intel store — and said nothing, because+ # nothing compared the two.+ #+ # Container ports are the *container's* ports from docker-compose.yml,+ # not the published host ports: 8002:8085 means actions listens on 8085+ # and compose maps it to 8002 on the host. Taking the left-hand number+ # is how a chart renders cleanly and every probe fails.+ actions:+ image:+ repository: ghcr.io/beenuar/aisoc-actions+ tag: "" # defaults to Chart.AppVersion+ pullPolicy: IfNotPresent+ replicaCount: 2+ containerPort: 8085+ healthCheckPath: /health+ readinessPath: /health+ service:+ type: ClusterIP+ port: 80+ resources:+ requests:+ cpu: 100m+ memory: 256Mi+ limits:+ cpu: 500m+ memory: 512Mi+ hpa:+ enabled: false+ pdb:+ enabled: true+ minAvailable: 1+ env: {}+ extraVolumes: []+ extraVolumeMounts: []++ connectors:+ image:+ repository: ghcr.io/beenuar/aisoc-connectors+ tag: ""+ pullPolicy: IfNotPresent+ # One replica by default, and that is a correctness constraint rather+ # than a sizing choice until the scheduler's leader election is+ # enabled: two replicas each run every connector's poll job, so every+ # vendor is polled twice and the duplicate events are deduplicated+ # downstream at best. See AISOC_SCHEDULER_LEADER_ELECTION.+ replicaCount: 1+ containerPort: 8003+ healthCheckPath: /health+ readinessPath: /health+ service:+ type: ClusterIP+ port: 80+ resources:+ requests:+ cpu: 100m+ memory: 256Mi+ limits:+ cpu: 500m+ memory: 768Mi+ hpa:+ enabled: false+ pdb:+ enabled: false+ env:+ AISOC_SCHEDULER_LEADER_ELECTION: "true"+ extraVolumes: []+ extraVolumeMounts: []++ threatintel:+ image:+ repository: ghcr.io/beenuar/aisoc-threatintel+ tag: ""+ pullPolicy: IfNotPresent+ replicaCount: 1+ containerPort: 8005+ healthCheckPath: /health+ readinessPath: /health+ service:+ type: ClusterIP+ port: 80+ resources:+ requests:+ cpu: 100m+ memory: 256Mi+ limits:+ cpu: 500m+ memory: 512Mi+ hpa:+ enabled: false+ pdb:+ enabled: false+ env: {}+ extraVolumes: []+ extraVolumeMounts: []+ingress:enabled: trueclassName: nginx@@ -514,6 +612,94 @@ backup:# rolsuper and rolbypassrls must both read false and the role must own nothing.# Skip step 1 and the services cannot authenticate — a loud failure. Skip step 2# and they authenticate as the owner, which is the silent one.+# Depth plan 4.5 / parity 6.8 — the three CORE services that are not ours.+#+# Separate from `.Values.services` on purpose: everything in that map+# defaults its image tag to `.Chart.AppVersion`, and tagging a+# third-party image with AiSOC's version produces a chart that renders+# perfectly and pulls a tag that does not exist. Two of them also need+# storage, which the stateless Deployment template has nowhere to put.+llmGateway:+ enabled: true+ image:+ # Pinned, not floating. `main-stable` moves, and a gateway that+ # changes under a running deployment changes which models resolve.+ repository: ghcr.io/berriai/litellm+ tag: main-stable+ pullPolicy: IfNotPresent+ replicaCount: 1+ env:+ # Every value the gateway reads must be an environment variable, not+ # a literal in the config: a bare literal cannot be changed without+ # rebuilding, which is how a request timeout came to be unraisable.+ LITELLM_REQUEST_TIMEOUT: "600"+ resources:+ requests:+ cpu: 100m+ memory: 256Mi+ limits:+ cpu: 500m+ memory: 512Mi+ # Rendered into a ConfigMap. Kept minimal here; a deployment with its+ # own model routing replaces this wholesale.+ config: |+ model_list:+ - model_name: aisoc-triage+ litellm_params:+ model: ollama/llama3.2:3b-instruct-q4_K_M+ api_base: os.environ/OLLAMA_BASE_URL+ router_settings:+ # An int, not a string. LiteLLM passes router_settings through+ # with no coercion and evaluates `num_retries > 0`, which raises+ # TypeError against a string at exactly the moment a retry is+ # needed.+ num_retries: 2+ general_settings:+ request_timeout: os.environ/LITELLM_REQUEST_TIMEOUT++ollama:+ # On by default because CORE ships a local model so a first run can+ # triage with zero credentials. A deployment pointing at a hosted+ # provider turns this off and keeps the gateway.+ enabled: true+ image:+ repository: ollama/ollama+ tag: 0.6.7+ pullPolicy: IfNotPresent+ model: llama3.2:3b-instruct-q4_K_M+ env: {}+ # The first start pulls gigabytes. 90 x 10s is fifteen minutes, which+ # is a slow registry rather than a broken pod; a probe tuned for a+ # stateless service kills it mid-pull, forever.+ startupFailureThreshold: 90+ persistence:+ size: 20Gi+ storageClass: ""+ resources:+ requests:+ cpu: 500m+ memory: 4Gi+ limits:+ cpu: 4000m+ memory: 8Gi++qdrant:+ enabled: true+ image:+ repository: qdrant/qdrant+ tag: v1.7.0+ pullPolicy: IfNotPresent+ persistence:+ size: 10Gi+ storageClass: ""+ resources:+ requests:+ cpu: 100m+ memory: 512Mi+ limits:+ cpu: 1000m+ memory: 2Gi+postgresql:enabled: falseauth: