diff --git a/docs/internal/cloud-matrix.md b/docs/internal/cloud-matrix.md
index 2d92615a5..d17b43afe 100644
--- a/docs/internal/cloud-matrix.md
+++ b/docs/internal/cloud-matrix.md
@@ -355,10 +355,21 @@ Storage auth uses GCP Workload Identity via Application Default Credentials (ADC
 config:
   values:
     launch:
-      cache_redis_onprem_url: redis://my-memorystore-instance.redis.cache.googleapis.com:6379/0
+      cache_redis_gcp_url: redis://my-memorystore-instance.redis.cache.googleapis.com:6379/0
 ```
 
-Alternatively, `cache_redis_aws_url` is also accepted (the on-prem fallback path in `config.py` accepts it with a log warning). The same Redis instance serves as both the caching layer and the Celery broker.
+`cache_redis_gcp_url` is the preferred field for GCP — it maps directly to GCP Memorystore and triggers the GCP branch in `HostedModelInferenceServiceConfig.cache_redis_url`. The same Redis instance serves as both the caching layer and the Celery broker.
+
+#### Database (Cloud SQL)
+
+GCP uses environment variables for DB credentials injected via a Kubernetes secret — the same pattern as on-prem. There is no GCP Secret Manager integration; use `kubernetesDatabaseSecretName` to mount the secret:
+
+```yaml
+secrets:
+  kubernetesDatabaseSecretName: llm-engine-postgres-credentials
+```
+
+The K8s secret must contain keys that the chart injects as env vars: `DB_HOST`, `DB_HOST_RO`, `DB_PORT`, `DB_USER`, `DB_PASSWORD`, `DB_NAME`. `get_engine_url` in `db/base.py` reads these directly when `cloud_provider == "gcp"`.
 
 #### Service Account (GCP Workload Identity)
 
diff --git a/model-engine/model_engine_server/common/config.py b/model-engine/model_engine_server/common/config.py
index 902c1a898..9e6e39b81 100644
--- a/model-engine/model_engine_server/common/config.py
+++ b/model-engine/model_engine_server/common/config.py
@@ -77,6 +77,9 @@ class HostedModelInferenceServiceConfig:
         None  # Not an env var because the redis cache info is already here
     )
     cache_redis_onprem_url: Optional[str] = None  # For on-prem Redis (e.g., redis://redis:6379/0)
+    cache_redis_gcp_url: Optional[str] = (
+        None  # For GCP Memorystore (e.g., redis://MEMORYSTORE_HOST:6379/0)
+    )
     sglang_repository: Optional[str] = None
 
     @classmethod
@@ -106,6 +109,10 @@ def cache_redis_url(self) -> str:
             redis_port = getattr(infra_config(), "redis_port", 6379)
             return f"redis://{redis_host}:{redis_port}/0"
 
+        if cloud_provider == "gcp":
+            assert self.cache_redis_gcp_url, "cache_redis_gcp_url required for GCP"
+            return self.cache_redis_gcp_url
+
         if self.cache_redis_aws_url:
             assert cloud_provider == "aws", "cache_redis_aws_url is only for AWS"
             if self.cache_redis_aws_secret_name:
diff --git a/model-engine/model_engine_server/db/base.py b/model-engine/model_engine_server/db/base.py
index f5ea49e7c..3d2f884a3 100644
--- a/model-engine/model_engine_server/db/base.py
+++ b/model-engine/model_engine_server/db/base.py
@@ -69,6 +69,20 @@ def get_engine_url(
 
             engine_url = f"postgresql://{user}:{password}@{host}:{port}/{dbname}"
 
+        elif infra_config().cloud_provider == "gcp":
+            user = os.environ.get("DB_USER", "postgres")
+            password = os.environ.get("DB_PASSWORD", "postgres")
+            host = (
+                os.environ.get("DB_HOST_RO")
+                if read_only
+                else os.environ.get("DB_HOST", "localhost")
+            )
+            port = os.environ.get("DB_PORT", "5432")
+            dbname = os.environ.get("DB_NAME", "llm_engine")
+            logger.info(f"Connecting to db {host}:{port}, name {dbname}")
+
+            engine_url = f"postgresql://{user}:{password}@{host}:{port}/{dbname}"
+
         elif infra_config().cloud_provider == "azure":
             client = SecretClient(
                 vault_url=f"https://{os.environ.get('KEYVAULT_NAME')}.vault.azure.net",
diff --git a/model-engine/model_engine_server/inference/vllm/build_and_upload_image.sh b/model-engine/model_engine_server/inference/vllm/build_and_upload_image.sh
index 82f2b9567..ea7124c24 100755
--- a/model-engine/model_engine_server/inference/vllm/build_and_upload_image.sh
+++ b/model-engine/model_engine_server/inference/vllm/build_and_upload_image.sh
@@ -38,7 +38,8 @@ set -eo pipefail
 # Examples:
 #
 #   # 1. Published versions (base image and pip version match)
-#   ./build_and_upload_image.sh my-tag vllm
+#   ./build_and_upload_image.sh my-tag vllm                          # → tag: 0.17.0-my-tag
+#   ./build_and_upload_image.sh "" vllm                              # → tag: 0.17.0 (version-only)
 #   ./build_and_upload_image.sh my-tag vllm --vllm-version=0.15.1
 #
 #   # 2. Newer pip version on older base image (e.g. 0.16.0 wheel on 0.15.1 base)
@@ -105,8 +106,8 @@ VLLM_OMNI_VERSION=${VLLM_OMNI_VERSION:-"0.16.0"}
 VLLM_OMNI_SOURCE_DIR=""
 VLLM_OMNI_SOURCE_REF=""
 
-if [ -z "$1" ]; then
-  echo "Must supply the user-provided tag"
+if [ "$#" -lt 1 ]; then
+  echo "Must supply the user-provided tag (pass empty string \"\" for a version-only tag)"
   exit 1;
 fi
 
@@ -264,9 +265,9 @@ fi
 
 # Construct image tag based on vllm version and user tag
 if [ "$BUILD_TARGET" == "vllm_omni" ]; then
-  IMAGE_TAG="${VLLM_VERSION}-omni-${USER_TAG}"
+  IMAGE_TAG="${VLLM_VERSION}-omni${USER_TAG:+-$USER_TAG}"
 else
-  IMAGE_TAG="${VLLM_VERSION}-${USER_TAG}"
+  IMAGE_TAG="${VLLM_VERSION}${USER_TAG:+-$USER_TAG}"
 fi
 
 # if build target = vllm use vllm otherwise use vllm_batch
diff --git a/model-engine/tests/unit/api/test_dependencies.py b/model-engine/tests/unit/api/test_dependencies.py
index 15491c368..a2712827e 100644
--- a/model-engine/tests/unit/api/test_dependencies.py
+++ b/model-engine/tests/unit/api/test_dependencies.py
@@ -1,6 +1,8 @@
 from unittest.mock import MagicMock, patch
 
+import pytest
 from model_engine_server.api.dependencies import _get_external_interfaces
+from model_engine_server.common.config import HostedModelInferenceServiceConfig
 from model_engine_server.infra.gateways import (
     GCSFileStorageGateway,
     GCSFilesystemGateway,
@@ -119,3 +121,71 @@ def test_gcp_provider_selects_gcp_implementations():
             external_interfaces.resource_gateway.queue_delegate,
             RedisQueueEndpointResourceDelegate,
         )
+
+
+def test_gcp_cache_redis_url_returns_gcp_url():
+    """Test that cache_redis_url returns cache_redis_gcp_url when cloud_provider is gcp."""
+    with patch("model_engine_server.common.config.infra_config") as mock_infra_config:
+        mock_infra = MagicMock()
+        mock_infra.cloud_provider = "gcp"
+        mock_infra_config.return_value = mock_infra
+
+        config = HostedModelInferenceServiceConfig(
+            gateway_namespace="ns",
+            endpoint_namespace="ns",
+            billing_queue_arn="arn",
+            sqs_profile="default",
+            sqs_queue_policy_template="{}",
+            sqs_queue_tag_template="{}",
+            model_primitive_host="localhost",
+            cloud_file_llm_fine_tune_repository="gs://bucket/ft",
+            hf_user_fine_tuned_weights_prefix="gs://bucket/weights",
+            istio_enabled=False,
+            dd_trace_enabled=False,
+            tgi_repository="repo/tgi",
+            vllm_repository="repo/vllm",
+            lightllm_repository="repo/lightllm",
+            tensorrt_llm_repository="repo/tensorrt",
+            batch_inference_vllm_repository="repo/batch",
+            user_inference_base_repository="repo/base",
+            user_inference_pytorch_repository="repo/pytorch",
+            user_inference_tensorflow_repository="repo/tf",
+            docker_image_layer_cache_repository="repo/cache",
+            sensitive_log_mode=False,
+            cache_redis_gcp_url="redis://10.0.0.1:6379/0",
+        )
+        assert config.cache_redis_url == "redis://10.0.0.1:6379/0"
+
+
+def test_gcp_cache_redis_url_raises_when_not_set():
+    """Test that cache_redis_url raises AssertionError for GCP when cache_redis_gcp_url is not set."""
+    with patch("model_engine_server.common.config.infra_config") as mock_infra_config:
+        mock_infra = MagicMock()
+        mock_infra.cloud_provider = "gcp"
+        mock_infra_config.return_value = mock_infra
+
+        config = HostedModelInferenceServiceConfig(
+            gateway_namespace="ns",
+            endpoint_namespace="ns",
+            billing_queue_arn="arn",
+            sqs_profile="default",
+            sqs_queue_policy_template="{}",
+            sqs_queue_tag_template="{}",
+            model_primitive_host="localhost",
+            cloud_file_llm_fine_tune_repository="gs://bucket/ft",
+            hf_user_fine_tuned_weights_prefix="gs://bucket/weights",
+            istio_enabled=False,
+            dd_trace_enabled=False,
+            tgi_repository="repo/tgi",
+            vllm_repository="repo/vllm",
+            lightllm_repository="repo/lightllm",
+            tensorrt_llm_repository="repo/tensorrt",
+            batch_inference_vllm_repository="repo/batch",
+            user_inference_base_repository="repo/base",
+            user_inference_pytorch_repository="repo/pytorch",
+            user_inference_tensorflow_repository="repo/tf",
+            docker_image_layer_cache_repository="repo/cache",
+            sensitive_log_mode=False,
+        )
+        with pytest.raises(AssertionError, match="cache_redis_gcp_url required for GCP"):
+            _ = config.cache_redis_url