Skip to content

armgit / AiSOC

Public Imported from GitHub
0

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.

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.yaml
new file mode 100644
index 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.yaml
index 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: true
className: 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: false
auth: