Use local OKP embeddings from rag-content

The rag-content image now ships the OKP embedding model in it and we no
longer need for OGX to download it at runtime via HF.

Signed-off-by: Lucas Alvares Gomes <lucasagomes@gmail.com>

Author: umago

api/v1beta1/openstacklightspeed_types.go

@@ -29,7 +29,7 @@ const (
 	// the automated pipeline for building OGX-compatible vector database images
 	// is ready.
 	// OpenStackLightspeedContainerImage is the fall-back container image for OpenStackLightspeed
-	OpenStackLightspeedContainerImage = "quay.io/openstack-lightspeed/rag-content:alpha-ogx-os-docs-2025.2"
+	OpenStackLightspeedContainerImage = "quay.io/openstack-lightspeed/rag-content:os-docs-2026.1-ogx"
 
 	// LCoreContainerImage is the fall-back container image for LCore
 	LCoreContainerImage = "quay.io/lightspeed-core/lightspeed-stack:dev-latest"

bundle/manifests/openstack-lightspeed-operator.clusterserviceversion.yaml

@@ -25,7 +25,7 @@ metadata:
       ]
     capabilities: Basic Install
     categories: AI/Machine Learning
-    createdAt: "2026-06-11T09:06:59Z"
+    createdAt: "2026-06-19T15:37:35Z"
     description: AI-powered virtual assistant for Red Hat OpenStack Services on OpenShift
     features.operators.openshift.io/cnf: "false"
     features.operators.openshift.io/cni: "false"
@@ -292,7 +292,7 @@ spec:
                     fieldRef:
                       fieldPath: metadata.annotations['olm.targetNamespaces']
                 - name: RELATED_IMAGE_OPENSTACK_LIGHTSPEED_IMAGE_URL_DEFAULT
-                  value: quay.io/openstack-lightspeed/rag-content:alpha-ogx-os-docs-2025.2
+                  value: quay.io/openstack-lightspeed/rag-content:os-docs-2026.1-ogx
                 - name: RELATED_IMAGE_LCORE_IMAGE_URL_DEFAULT
                   value: quay.io/lightspeed-core/lightspeed-stack:dev-latest
                 - name: RELATED_IMAGE_EXPORTER_IMAGE_URL_DEFAULT
@@ -474,7 +474,7 @@ spec:
     name: Red Hat
     url: https://github.com/openstack-lightspeed/operator
   relatedImages:
-  - image: quay.io/openstack-lightspeed/rag-content:alpha-ogx-os-docs-2025.2
+  - image: quay.io/openstack-lightspeed/rag-content:os-docs-2026.1-ogx
     name: openstack-lightspeed-image-url-default
   - image: quay.io/lightspeed-core/lightspeed-stack:dev-latest
     name: lcore-image-url-default

config/manager/manager.yaml

@@ -74,7 +74,7 @@ spec:
             # the automated pipeline for building OGX-compatible vector database images
             # is ready.
           - name: RELATED_IMAGE_OPENSTACK_LIGHTSPEED_IMAGE_URL_DEFAULT
-            value: quay.io/openstack-lightspeed/rag-content:alpha-ogx-os-docs-2025.2
+            value: quay.io/openstack-lightspeed/rag-content:os-docs-2026.1-ogx
           - name: RELATED_IMAGE_LCORE_IMAGE_URL_DEFAULT
             value: quay.io/lightspeed-core/lightspeed-stack:dev-latest
           - name: RELATED_IMAGE_EXPORTER_IMAGE_URL_DEFAULT

config/manifests/bases/openstack-lightspeed-operator.clusterserviceversion.yaml

@@ -171,6 +171,6 @@ spec:
   relatedImages:
   - image: quay.io/openstack-lightspeed/operator:latest
     name: operator
-  - image: quay.io/openstack-lightspeed/rag-content:os-docs-2025.2
+  - image: quay.io/openstack-lightspeed/rag-content:os-docs-2026.1-ogx
     name: rag-content
   version: 0.0.0

hack/env.sh

@@ -5,6 +5,6 @@ export RELATED_IMAGE_POSTGRES_IMAGE_URL_DEFAULT="registry.redhat.io/rhel9/postgr
 # TODO(lpiwowar): Replace this with a stable (non-alpha) image version once
 # the automated pipeline for building OGX-compatible vector database images
 # is ready.
-export RELATED_IMAGE_OPENSTACK_LIGHTSPEED_IMAGE_URL_DEFAULT="quay.io/openstack-lightspeed/rag-content:alpha-ogx-os-docs-2025.2"
+export RELATED_IMAGE_OPENSTACK_LIGHTSPEED_IMAGE_URL_DEFAULT="quay.io/openstack-lightspeed/rag-content:os-docs-2026.1-ogx"
 export RELATED_IMAGE_OKP_IMAGE_URL_DEFAULT="registry.redhat.io/offline-knowledge-portal/rhokp-rhel9:latest"
 export WATCH_NAMESPACE="openstack-lightspeed"

internal/controller/assets/vector_database_collect.sh

@@ -38,18 +38,22 @@
 # │   └── ocp_latest/
 # │       ├── faiss_store.db
 # │       └── llama-stack.yaml
-# └── embeddings_model/
+# ├── embeddings_model/
+# │   └── <model_files>
+# └── okp_embeddings_model/
 #     └── <model_files>
 #
 # Output Structure:
 # <target-path>/            (specified via --vector-db-path)
-# └── <random-tmp-dir>/
-#     ├── vector_db/
-#     │   ├── vector-db-data-1/
-#     │   ├── vector-db-data-N/
-#     │   └── ocp_X.YZ/     (if --enable-ocp-rag true and --ocp-version X.YZ)
-#     └── embeddings_model/
-#         └── <model_files>
+# ├── <random-tmp-dir>/
+# │   ├── vector_db/
+# │   │   ├── vector-db-data-1/
+# │   │   ├── vector-db-data-N/
+# │   │   └── ocp_X.YZ/     (if --enable-ocp-rag true and --ocp-version X.YZ)
+# │   └── embeddings_model/
+# │       └── <model_files>
+# └── okp_embeddings_model/  (if present in the image)
+#     └── <model_files>
 #
 # Arguments:
 #  --vector-db-path PATH    Target directory for collected data (required)
@@ -110,6 +114,10 @@ VECTOR_DB_DIR="/rag/vector_db"
 # where embeddings model must reside.
 EMBEDDINGS_MODEL_DIR="/rag/embeddings_model"
 
+# OKP_EMBEDDING_MODEL_SRC specifies the directory within the vector DB container
+# image where the OKP embedding model must reside.
+OKP_EMBEDDING_MODEL_SRC="/rag/okp_embeddings_model"
+
 # OGX_CONFIG_FILE_NAME is the name of the OGX config file associated with a
 # single vector database.
 OGX_CONFIG_FILE_NAME="llama-stack.yaml"
@@ -246,13 +254,26 @@ collect_embeddings_model() {
     echo "Discovered and collected embeddings model data from ${EMBEDDINGS_MODEL_DIR}"
 }
 
+collect_okp_embeddings_model() {
+    if [ ! -d "${OKP_EMBEDDING_MODEL_SRC}" ]; then
+        echo "OKP embedding model not found under ${OKP_EMBEDDING_MODEL_SRC} => Skipping"
+        return
+    fi
+
+    echo "Collecting OKP embedding model ..."
+    rm -rf "${VECTOR_DB_VOLUME_MOUNT_PATH}/okp_embeddings_model"
+    cp -rL "${OKP_EMBEDDING_MODEL_SRC}" "${VECTOR_DB_VOLUME_MOUNT_PATH}/okp_embeddings_model"
+    echo "Discovered and collected OKP embedding model from ${OKP_EMBEDDING_MODEL_SRC}"
+}
+
 main() {
     # NOTE: parse_arguments_and_init must be called first to ensure all global
     # variables are initialized before proceeding.
     parse_arguments_and_init "$@"
     collect_vector_db_data
     collect_ocp_vector_db_data
     collect_embeddings_model
+    collect_okp_embeddings_model
 }
 
 main "$@"

internal/controller/constants.go

@@ -134,6 +134,11 @@ const (
 	// init container responsible for assembling the final Lightspeed Stack config.
 	VectorDBVolumeLightspeedStackConfigPath = VectorDBVolumeMountPath + "/lightspeed-stack.yaml"
 
+	// OKPEmbeddingModelMountPath specifies the path within the VectorDBVolumeName
+	// volume where the OKP embedding model is stored after being extracted from the
+	// rag-content image by the vector-database-collect init container.
+	OKPEmbeddingModelMountPath = VectorDBVolumeMountPath + "/okp_embeddings_model"
+
 	// OGXConfigInitContainerMountPath specifies the path where the operator-generated
 	// OGX config file is mounted in the init container responsible for assembling
 	// the final OGX configuration, which includes information about RAG.

internal/controller/lcore_deployment.go

@@ -570,13 +570,6 @@ func buildLlamaStackEnvVars(h *common_helper.Helper, ctx context.Context, instan
 			Name:  "RH_SERVER_OKP",
 			Value: fmt.Sprintf("http://%s.%s.svc:%d", OKPServiceName, instance.GetNamespace(), OKPServicePort),
 		})
-		// FIXME(lucasagomes): Llama-Stack expects HF_HOME to be set when OKP is enabled because it uses the
-		// Hugging Face Hub client to fetch the embedding model for OKP. Ideally we would include the model it
-		// downloads in the container image to avoid this.
-		envVars = append(envVars, corev1.EnvVar{
-			Name:  "HF_HOME",
-			Value: "/tmp/huggingface",
-		})
 	}
 
 	return envVars, nil

internal/controller/llama_stack_config.go

@@ -257,7 +257,7 @@ func buildOKPVectorIOProvider(instance *apiv1beta1.OpenStackLightspeed) map[stri
 			"content_field":       "${env.SOLR_CONTENT_FIELD:=chunk}",
 			"vector_field":        "${env.SOLR_VECTOR_FIELD:=chunk_vector}",
 			"embedding_dimension": "${env.SOLR_EMBEDDING_DIM:=384}",
-			"embedding_model":     "${env.SOLR_EMBEDDING_MODEL:=sentence-transformers/ibm-granite/granite-embedding-30m-english}",
+			"embedding_model":     "sentence-transformers" + OKPEmbeddingModelMountPath,
 			"persistence": map[string]interface{}{
 				"backend":   "kv_default",
 				"namespace": "portal-rag",
@@ -366,7 +366,7 @@ func buildLlamaStackModels(_ *common_helper.Helper, instance *apiv1beta1.OpenSta
 			"model_id":          "solr_embedding",
 			"model_type":        "embedding",
 			"provider_id":       "sentence-transformers",
-			"provider_model_id": "${env.SOLR_EMBEDDING_MODEL:=ibm-granite/granite-embedding-30m-english}",
+			"provider_model_id": OKPEmbeddingModelMountPath,
 			"metadata": map[string]interface{}{
 				"embedding_dimension": 384,
 			},
@@ -383,7 +383,7 @@ func buildLlamaStackVectorStores(_ *common_helper.Helper, instance *apiv1beta1.O
 			"vector_store_id":     "portal-rag",
 			"provider_id":         "okp_solr",
 			"embedding_dimension": 384,
-			"embedding_model":     "${env.SOLR_EMBEDDING_MODEL:=sentence-transformers/ibm-granite/granite-embedding-30m-english}",
+			"embedding_model":     "sentence-transformers" + OKPEmbeddingModelMountPath,
 		})
 	}
 	return stores

kuttl-test.yaml

@@ -6,5 +6,3 @@ reportGranularity: test
 namespace: openstack-lightspeed
 timeout: 1380
 parallel: 1
-suppress:
-  - events