Compare commits

...
Author SHA1 Message Date
claude-bot fd2f3b7be4 fix(hermes): keep image automation off the retired date-versioned tags
review-bot reviewed: comment
ci/woodpecker/pr/ci-scripts-test Pipeline was successful
ci/woodpecker/pr/ci-manifests-test Pipeline was successful
nousresearch/hermes-agent still carries v2026.4.3..v2026.9.24 from the
old date-based scheme. As semver they sort above 0.x, so '>=0.22.2'
picked v2026.9.24 (2026-09-24) and rolled the pod back from v0.22.2.
Cap the range below 1.0.0 and pin the manifests back to v0.22.2.
2026-10-19 08:18:03 +02:00
k.renner e614c748b9 Merge pull request 'fix(kuma-sync): create when the previous monitor ID belongs to another owner' (#1354) from fix/kuma-sync-foreign-key-create into master
ci/woodpecker/cron/alert-monitor Pipeline was successful
ci/woodpecker/push/build-ci-images Pipeline was successful
Reviewed-on: k.renner/homelab#1354
2026-10-19 06:14:48 +00:00
security-bot 30b0f43911 chore(image-automation): bump nousresearch/hermes-agent:v2026.9.24
Updates managed by image-automation-controller.
2026-10-19 06:13:48 +00:00
k.renner 10819fac62 Merge pull request 'feat(hermes): run Hermes Agent in the cluster' (#1352) from feat/hermes-agent into master
Reviewed-on: k.renner/homelab#1352
2026-10-19 06:13:17 +00:00
claude-bot dd954ce23a feat(hermes): run Hermes Agent in the cluster
ci/woodpecker/pr/ci-scripts-test Pipeline was successful
ci/woodpecker/pr/ci-manifests-test Pipeline was successful
review-bot reviewed: request changes
Hermes Agent as a pod in its own `hermes` namespace: gateway (messaging
platforms, cron, OpenAI-compatible API) and dashboard, supervised by the
image's s6-overlay, state on a 10Gi ceph-block PVC.

- hermes.renner.uno (LAN): dashboard and Hermes Desktop remote gateway.
  Plain Traefik plus the dashboard's own Authentik OIDC login (public PKCE
  client, Hermes Users group) — a forward-auth gate would break Desktop's
  Bearer-token calls.
- hermes-api.renner.uno (LAN): API server, Bearer key generated in-cluster
  by an ESO Password generator.
- Read-only kubectl via the built-in `view` ClusterRole.
- CNPs: world 443/80, kube-apiserver, Traefik for LAN hostnames, and
  Prometheus/Alertmanager/Loki (with the matching monitoring-side grant).
- Image automation for nousresearch/hermes-agent v-tags.

After deploy: log in to Anthropic inside the pod
(`hermes auth add anthropic --type oauth --no-browser`, then `hermes model`);
platform and Home Assistant tokens go in via the dashboard.
Not yet covered: a Velero schedule for the PVC.
2026-10-19 08:02:56 +02:00
13 changed files with 575 additions and 0 deletions
@@ -974,6 +974,71 @@ data:
negate: false
failure_result: false
# Hermes Agent dashboard (apps/hermes): its self-hosted OIDC login,
# also brokered for Hermes Desktop's native sign-in. Public client +
# PKCE (Hermes does not support confidential clients); RS256 so the
# dashboard can verify the ID token against the JWKS.
- id: hermes-users-group
model: authentik_core.group
identifiers:
name: Hermes Users
state: present
attrs:
is_superuser: false
# Members get the agent, which holds a read-only kubeconfig and
# the model login. Add/remove members HERE.
users:
- !Find [authentik_core.user, [username, k.renner]]
- id: hermes-provider
model: authentik_providers_oauth2.oauth2provider
identifiers:
name: Hermes Dashboard
state: present
attrs:
authentication_flow: null
authorization_flow: !Find [authentik_flows.flow, [slug, default-provider-authorization-implicit-consent]]
invalidation_flow: !Find [authentik_flows.flow, [slug, default-provider-invalidation-flow]]
client_type: public
client_id: hermes-dashboard
# See the Matrix (MAS) provider: a blueprint-made provider starts
# with no grant types.
grant_types:
- authorization_code
- refresh_token
signing_key: !Find [authentik_crypto.certificatekeypair, [name, authentik Self-signed Certificate]]
property_mappings:
- !Find [authentik_providers_oauth2.scopemapping, [managed, goauthentik.io/providers/oauth2/scope-openid]]
- !Find [authentik_providers_oauth2.scopemapping, [managed, goauthentik.io/providers/oauth2/scope-profile]]
- !Find [authentik_providers_oauth2.scopemapping, [managed, goauthentik.io/providers/oauth2/scope-email]]
redirect_uris:
- url: https://hermes.renner.uno/auth/callback
matching_mode: strict
# The slug is the issuer path: https://auth.renner.uno/application/o/hermes/
- model: authentik_core.application
identifiers:
slug: hermes
state: present
attrs:
name: Hermes
slug: hermes
provider: !KeyOf hermes-provider
meta_launch_url: https://hermes.renner.uno
meta_description: "Hermes Agent dashboard"
# Only the Hermes Users group passes.
- model: authentik_policies.policybinding
identifiers:
target: !Find [authentik_core.application, [slug, hermes]]
group: !KeyOf hermes-users-group
order: 0
state: present
attrs:
enabled: true
negate: false
failure_result: false
# OpenProject, decommissioned. Application first: Application.provider
# is on_delete=PROTECT. Its policy binding cascades with it.
- model: authentik_core.application
+32
View File
@@ -0,0 +1,32 @@
# Bearer key for the OpenAI-compatible API server (hermes-api.renner.uno).
# Generated once in-cluster, nothing in OpenBao. Read it with
# kubectl -n hermes get secret hermes-api-key -o jsonpath='{.data.API_SERVER_KEY}' | base64 -d
apiVersion: generators.external-secrets.io/v1alpha1
kind: Password
metadata:
name: hermes-api-key
namespace: hermes
spec:
length: 64
digits: 16
symbols: 0
allowRepeat: true
noUpper: false
secretKeys:
- API_SERVER_KEY
---
apiVersion: external-secrets.io/v1
kind: ExternalSecret
metadata:
name: hermes-api-key
namespace: hermes
spec:
refreshPolicy: CreatedOnce
target:
name: hermes-api-key
dataFrom:
- sourceRef:
generatorRef:
apiVersion: generators.external-secrets.io/v1alpha1
kind: Password
name: hermes-api-key
+170
View File
@@ -0,0 +1,170 @@
# All Hermes state: config.yaml, .env (keys entered in the dashboard), auth.json
# (the Anthropic OAuth login), sessions, memories, skills, cron. Mounted at
# /opt/data, the image's HERMES_HOME.
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
name: hermes-data
namespace: hermes
spec:
accessModes:
- ReadWriteOnce
storageClassName: ceph-block
resources:
requests:
storage: 10Gi
---
# Hermes Agent: the gateway (messaging platforms, cron, API server on 8642) and
# the dashboard (9119, also the Hermes Desktop remote-gateway backend), both
# supervised by the image's s6-overlay. s6 must be PID 1 and starts as root,
# remapping and dropping to the `hermes` user (10000).
#
# One replica, Recreate: two gateways must never share one data directory.
apiVersion: apps/v1
kind: Deployment
metadata:
name: hermes
namespace: hermes
labels:
app.kubernetes.io/name: hermes
spec:
replicas: 1
strategy:
type: Recreate
selector:
matchLabels:
app.kubernetes.io/name: hermes
template:
metadata:
labels:
app.kubernetes.io/name: hermes
spec:
serviceAccountName: hermes
securityContext:
fsGroup: 10000
seccompProfile:
type: RuntimeDefault
initContainers:
# The dashboard trusts X-Forwarded-* only from listed peers; Traefik
# connects from the pod network. No env equivalent exists, so it is
# set in config.yaml on every start (idempotent).
- name: config
image: nousresearch/hermes-agent:v0.22.2 # {"$imagepolicy": "flux-system:hermes"}
command:
- /opt/hermes/.venv/bin/hermes
- config
- set
- dashboard.trusted_proxies
- '["10.244.0.0/16","fd10:244::/48"]'
env:
- name: HERMES_HOME
value: /opt/data
- name: HOME
value: /opt/data
securityContext:
runAsUser: 10000
runAsGroup: 10000
runAsNonRoot: true
allowPrivilegeEscalation: false
capabilities:
drop: ["ALL"]
volumeMounts:
- name: data
mountPath: /opt/data
# kubectl for the agent. /opt/data/.local/bin is on the image's PATH;
# an emptyDir there keeps the binary out of the PVC and current with
# this pin.
- name: kubectl
image: docker.io/alpine/kubectl:1.37.1@sha256:b568e8beb6a79aee269c142343078a278b85e7710f1c0e7d8d6b21d1eb566a48
command: ["cp", "/usr/local/bin/kubectl", "/tools/kubectl"]
securityContext:
runAsUser: 10000
runAsGroup: 10000
runAsNonRoot: true
allowPrivilegeEscalation: false
capabilities:
drop: ["ALL"]
volumeMounts:
- name: tools
mountPath: /tools
containers:
- name: hermes
image: nousresearch/hermes-agent:v0.22.2 # {"$imagepolicy": "flux-system:hermes"}
args: ["gateway", "run"]
env:
- name: HERMES_DASHBOARD
value: "1"
- name: HERMES_DASHBOARD_PUBLIC_URL
value: https://hermes.renner.uno
# Dashboard login: Authentik OIDC, public PKCE client (blueprint
# in apps/authentik/blueprints-configmap.yaml, group Hermes Users).
- name: HERMES_DASHBOARD_OIDC_ISSUER
value: https://auth.renner.uno/application/o/hermes/
- name: HERMES_DASHBOARD_OIDC_CLIENT_ID
value: hermes-dashboard
- name: API_SERVER_ENABLED
value: "true"
- name: API_SERVER_HOST
value: 0.0.0.0
envFrom:
- secretRef:
name: hermes-api-key
ports:
- name: dashboard
containerPort: 9119
- name: api
containerPort: 8642
# The dashboard is the user-facing surface; /api/status is public.
startupProbe:
httpGet:
path: /api/status
port: dashboard
periodSeconds: 10
failureThreshold: 30
readinessProbe:
httpGet:
path: /api/status
port: dashboard
periodSeconds: 15
livenessProbe:
httpGet:
path: /api/status
port: dashboard
periodSeconds: 60
timeoutSeconds: 10
failureThreshold: 3
resources:
requests:
cpu: 100m
memory: 512Mi
limits:
memory: 4Gi
volumeMounts:
- name: data
mountPath: /opt/data
- name: tools
mountPath: /opt/data/.local/bin
volumes:
- name: data
persistentVolumeClaim:
claimName: hermes-data
- name: tools
emptyDir: {}
---
apiVersion: v1
kind: Service
metadata:
name: hermes
namespace: hermes
labels:
app.kubernetes.io/name: hermes
spec:
selector:
app.kubernetes.io/name: hermes
ports:
- name: dashboard
port: 9119
targetPort: dashboard
- name: api
port: 8642
targetPort: api
+59
View File
@@ -0,0 +1,59 @@
# Both hosts are LAN-only (no dns.renner.uno/public or netbird.renner.uno/public
# label).
#
# hermes.renner.uno — dashboard and Hermes Desktop remote gateway. Plain
# Traefik, not the authentik-gate class: Hermes Desktop calls the API with a
# Bearer token from its native (PKCE) sign-in, which a forward-auth gate would
# redirect. Authentik still guards it: the dashboard's own OIDC login against
# the `hermes` application, restricted to the Hermes Users group.
apiVersion: networking.k8s.io/v1
kind: Ingress
metadata:
name: hermes
namespace: hermes
annotations:
cert-manager.io/cluster-issuer: letsencrypt-cluster
spec:
ingressClassName: traefik
tls:
- hosts:
- hermes.renner.uno
secretName: hermes-tls
rules:
- host: hermes.renner.uno
http:
paths:
- path: /
pathType: Prefix
backend:
service:
name: hermes
port:
number: 9119
---
# hermes-api.renner.uno — the OpenAI-compatible API server. Hermes checks
# `Authorization: Bearer <API_SERVER_KEY>` itself (secret hermes-api-key).
apiVersion: networking.k8s.io/v1
kind: Ingress
metadata:
name: hermes-api
namespace: hermes
annotations:
cert-manager.io/cluster-issuer: letsencrypt-cluster
spec:
ingressClassName: traefik
tls:
- hosts:
- hermes-api.renner.uno
secretName: hermes-api-tls
rules:
- host: hermes-api.renner.uno
http:
paths:
- path: /
pathType: Prefix
backend:
service:
name: hermes
port:
number: 8642
@@ -0,0 +1,7 @@
apiVersion: kustomize.config.k8s.io/v1beta1
kind: Kustomization
resources:
- rbac.yaml
- api-key.yaml
- hermes.yaml
- ingress.yaml
+22
View File
@@ -0,0 +1,22 @@
# Read-only cluster access for the agent's kubectl, as for claude-workstation:
# the built-in `view` ClusterRole, which excludes Secrets. The token is the
# pod's projected ServiceAccount token; kubectl picks it up as in-cluster
# config.
apiVersion: v1
kind: ServiceAccount
metadata:
name: hermes
namespace: hermes
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
name: hermes-view
roleRef:
apiGroup: rbac.authorization.k8s.io
kind: ClusterRole
name: view
subjects:
- kind: ServiceAccount
name: hermes
namespace: hermes
+18
View File
@@ -0,0 +1,18 @@
apiVersion: kustomize.toolkit.fluxcd.io/v1
kind: Kustomization
metadata:
name: hermes
namespace: flux-system
spec:
interval: 10m
retryInterval: 2m
timeout: 5m
sourceRef:
kind: GitRepository
name: flux-system
path: ./kubernetes/apps/hermes
prune: true
wait: true
dependsOn:
- name: infrastructure
- name: infrastructure-crds
@@ -0,0 +1,27 @@
apiVersion: image.toolkit.fluxcd.io/v1
kind: ImageRepository
metadata:
name: hermes
namespace: flux-system
spec:
image: nousresearch/hermes-agent
interval: 1h
---
# v-prefix tags (v0.22.2). The filter skips rc.*, -desktop and per-arch tags.
# The repo also carries the retired date-versioned releases (v2026.4.3 ..
# v2026.9.24, last one 2026-09-24); as semver they sort above 0.x, so the
# range caps below 1.0.0 or the policy picks the old v2026.9.24.
apiVersion: image.toolkit.fluxcd.io/v1
kind: ImagePolicy
metadata:
name: hermes
namespace: flux-system
spec:
imageRepositoryRef:
name: hermes
filterTags:
pattern: '^v(?P<ver>\d+\.\d+\.\d+)$'
extract: '$ver'
policy:
semver:
range: ">=0.22.2 <1.0.0"
@@ -10,6 +10,7 @@ resources:
- gatus.yaml
- gitea-mcp.yaml
- gotenberg.yaml
- hermes.yaml
- home-assistant.yaml
- immich.yaml
- kuma-sync.yaml
@@ -294,3 +294,12 @@ apiVersion: v1
kind: Namespace
metadata:
name: kuma-sync
---
# Hermes Agent (kubernetes/apps/hermes). Declared here so it exists before
# infrastructure-crds applies its CNPs. No pod-security labels: the image's
# s6-overlay init starts as root and drops to the `hermes` user, which the
# cluster's default `baseline` enforcement allows.
apiVersion: v1
kind: Namespace
metadata:
name: hermes
@@ -0,0 +1,137 @@
---
# Baseline policy for the `hermes` namespace (Hermes Agent, apps/hermes).
apiVersion: cilium.io/v2
kind: CiliumNetworkPolicy
metadata:
name: default-deny
namespace: hermes
spec:
endpointSelector: {}
ingress:
- fromEndpoints:
- matchLabels:
"k8s:io.kubernetes.pod.namespace": hermes
- fromEndpoints:
- matchLabels:
"k8s:io.kubernetes.pod.namespace": monitoring
- fromEntities:
- host
- remote-node
- fromEndpoints:
- matchExpressions:
- key: io.kubernetes.pod.namespace
operator: Exists
icmps:
- fields:
- { type: 0, family: IPv4 }
- { type: 3, family: IPv4 }
- { type: 8, family: IPv4 }
- { type: 11, family: IPv4 }
- { type: 12, family: IPv4 }
- { type: 1, family: IPv6 }
- { type: 2, family: IPv6 }
- { type: 3, family: IPv6 }
- { type: 4, family: IPv6 }
- { type: 128, family: IPv6 }
- { type: 129, family: IPv6 }
- { type: 133, family: IPv6 }
- { type: 134, family: IPv6 }
- { type: 135, family: IPv6 }
- { type: 136, family: IPv6 }
- { type: 143, family: IPv6 }
egress:
- toEndpoints:
- matchLabels:
"k8s:io.kubernetes.pod.namespace": hermes
- toEndpoints:
- matchLabels:
"k8s:io.kubernetes.pod.namespace": kube-system
k8s-app: kube-dns
toPorts:
- ports:
- port: "53"
protocol: UDP
- port: "53"
protocol: TCP
# Every LAN hostname (gitea, home, auth for the dashboard's OIDC
# discovery, matrix) via Traefik.
- toEndpoints:
- matchLabels:
"k8s:io.kubernetes.pod.namespace": ingress
app.kubernetes.io/name: traefik
---
# The agent itself: model APIs (api.anthropic.com), web tools, messaging
# platforms, read-only kubectl, and the monitoring read APIs. The matching
# ingress grant on the monitoring side is monitoring-ns.yaml
# (monitoring-allow-hermes).
apiVersion: cilium.io/v2
kind: CiliumNetworkPolicy
metadata:
name: hermes-egress
namespace: hermes
spec:
endpointSelector:
matchLabels:
app.kubernetes.io/name: hermes
egress:
- toEntities:
- world
toPorts:
- ports:
- port: "443"
protocol: TCP
- port: "80"
protocol: TCP
# `kube-apiserver` is its own entity in Cilium; port is the post-DNAT 6443.
- toEntities:
- kube-apiserver
toPorts:
- ports:
- port: "6443"
protocol: TCP
- toEndpoints:
- matchLabels:
"k8s:io.kubernetes.pod.namespace": monitoring
app.kubernetes.io/name: prometheus
toPorts:
- ports:
- port: "9090"
protocol: TCP
- toEndpoints:
- matchLabels:
"k8s:io.kubernetes.pod.namespace": monitoring
app.kubernetes.io/name: alertmanager
toPorts:
- ports:
- port: "9093"
protocol: TCP
- toEndpoints:
- matchLabels:
"k8s:io.kubernetes.pod.namespace": monitoring
app.kubernetes.io/name: loki
toPorts:
- ports:
- port: "3100"
protocol: TCP
---
# Traefik serves the dashboard (9119) and the OpenAI-compatible API (8642).
apiVersion: cilium.io/v2
kind: CiliumNetworkPolicy
metadata:
name: hermes-allow-traefik
namespace: hermes
spec:
endpointSelector:
matchLabels:
app.kubernetes.io/name: hermes
ingress:
- fromEndpoints:
- matchLabels:
"k8s:io.kubernetes.pod.namespace": ingress
app.kubernetes.io/name: traefik
toPorts:
- ports:
- port: "9119"
protocol: TCP
- port: "8642"
protocol: TCP
@@ -34,5 +34,6 @@ resources:
- uptime-monitor-ns.yaml
- kuma-sync-ns.yaml
- renovate-ns.yaml
- hermes-ns.yaml
- allow-cilium-health.yaml
- kube-system-hubble-relay.yaml
@@ -57,3 +57,30 @@ spec:
- { type: 135, family: IPv6 }
- { type: 136, family: IPv6 }
- { type: 143, family: IPv6 }
---
# Hermes Agent (hermes ns) reads metrics, alerts and logs: the Prometheus,
# Alertmanager and Loki HTTP APIs, scoped to those pods and ports.
apiVersion: cilium.io/v2
kind: CiliumNetworkPolicy
metadata:
name: monitoring-allow-hermes
namespace: monitoring
spec:
endpointSelector:
matchExpressions:
- key: app.kubernetes.io/name
operator: In
values: [prometheus, alertmanager, loki]
ingress:
- fromEndpoints:
- matchLabels:
"k8s:io.kubernetes.pod.namespace": hermes
app.kubernetes.io/name: hermes
toPorts:
- ports:
- port: "9090"
protocol: TCP
- port: "9093"
protocol: TCP
- port: "3100"
protocol: TCP