diff --git a/deploy/stacks/self-managed/.gitignore b/deploy/stacks/self-managed/.gitignore
new file mode 100644
index 000000000..89f9ac04a
--- /dev/null
+++ b/deploy/stacks/self-managed/.gitignore
@@ -0,0 +1 @@
+out/
diff --git a/tests/bdd/features/single-cluster-helmfile.feature b/tests/bdd/features/single-cluster-helmfile.feature
index 7efeccd26..a85fdbe7c 100644
--- a/tests/bdd/features/single-cluster-helmfile.feature
+++ b/tests/bdd/features/single-cluster-helmfile.feature
@@ -20,10 +20,11 @@ Feature: Install a local single-cluster NVCF stack with Helmfile
       # ingress.gatewayApi.*). The Background only overlays the operator-specific
       # values that vary per NGC org and pull-secret name.
       And I update yaml file "deploy/stacks/self-managed/environments/local-bdd.yaml" with keys:
-        | global.imagePullSecrets[0].name | nvcr-pull-secret                     |
-        | global.helm.sources.repository  | ${SAMPLE_NGC_ORG}/${SAMPLE_NGC_TEAM} |
-        | global.image.repository         | ${SAMPLE_NGC_ORG}/${SAMPLE_NGC_TEAM} |
-        | observability.profile           | disabled                             |
+        | global.imagePullSecrets[0].name               | nvcr-pull-secret                                                   |
+        | global.helm.sources.repository                | ${SAMPLE_NGC_ORG}/${SAMPLE_NGC_TEAM}                               |
+        | global.image.repository                       | ${SAMPLE_NGC_ORG}/${SAMPLE_NGC_TEAM}                               |
+        | api.env.NVCF_SIDECARS_LLM_ROUTER_CLIENT_IMAGE | nvcr.io/${SAMPLE_NGC_ORG}/${SAMPLE_NGC_TEAM}/stargate-client:0.2.0 |
+        | observability.profile                         | disabled                                                           |
       And I copy the file "tests/bdd/fixtures/nvcf-compute-plane-local-bdd.yaml" to "deploy/stacks/nvcf-compute-plane/environments/local-bdd.yaml"
       And I update yaml file "deploy/stacks/nvcf-compute-plane/environments/local-bdd.yaml" with keys:
         | global.imagePullSecrets[0].name | nvcr-pull-secret                     |
@@ -180,6 +181,14 @@ Feature: Install a local single-cluster NVCF stack with Helmfile
       Then the command exit code should be 0
       And the command output should contain "bdd-echo"
 
+      # Remove the deployment: the local sizing cannot hold every
+      # scenario's deployment at once.
+      When I run command:
+        """
+        ${NVCF_CLI} --config ${REPO_ROOT}/tests/bdd/fixtures/nvcf-cli-local.yaml function delete --deployment-only
+        """
+      Then the command exit code should be 0
+
     @function-lifecycle @grpc
     Scenario: Operator creates, deploys, and invokes the gRPC Load Tester Supreme sample function
       When I run command:
@@ -206,3 +215,62 @@ Feature: Install a local single-cluster NVCF stack with Helmfile
         """
       Then the command exit code should be 0
       And the command output should contain "bdd-grpc-echo"
+
+      # Free the GPU node for the LLM scenario.
+      When I run command:
+        """
+        ${NVCF_CLI} --config ${REPO_ROOT}/tests/bdd/fixtures/nvcf-cli-local.yaml function delete --deployment-only
+        """
+      Then the command exit code should be 0
+
+    # Proves the serve path the @llm-gateway scenario only installs:
+    # LLM-type functions route through llm-api-gateway and
+    # llm-request-router. The CLI maps invocation.localhost to the
+    # llm.localhost gateway host and rewrites the body model to
+    # <functionId>/<model>. Depends on the earlier install and
+    # registration scenarios; not a standalone tag target.
+    @llm-function-type
+    Scenario: Operator creates, deploys, and invokes an LLM-type OpenAI-compatible sample function
+      When I run command:
+        """
+        ${NVCF_CLI} --config ${REPO_ROOT}/tests/bdd/fixtures/nvcf-cli-local.yaml function create --name bdd-openai-compatible-sample --image nvcr.io/${SAMPLE_NGC_ORG}/${SAMPLE_NGC_TEAM}/nvcf-openai-compatible-sample:local --function-type LLM --inference-url /v1/chat/completions --inference-port 8000 --health-uri /health --health-port 8000 --health-timeout PT30S --llm-model 'name=openai-compatible-sample,uris=/v1/chat/completions|/v1/embeddings,routingMethod=round_robin'
+        """
+      Then the command exit code should be 0
+
+      When I run command:
+        """
+        ${NVCF_CLI} --config ${REPO_ROOT}/tests/bdd/fixtures/nvcf-cli-local.yaml function deploy create --gpu H100 --instance-type NCP.GPU.H100_8x --backend ncp-local --regions us-west-1 --min-instances 1 --max-instances 1 --timeout 900
+        """
+      Then the command exit code should be 0
+
+      When I run command:
+        """
+        ${NVCF_CLI} --config ${REPO_ROOT}/tests/bdd/fixtures/nvcf-cli-local.yaml api-key generate --description bdd-openai-compatible-sample --for function --scopes invoke_function,list_functions,queue_details,list_functions_details
+        """
+      Then the command exit code should be 0
+
+      # The gateway answers synchronously (no queue polling). The
+      # sample always returns its fixed load-testing message.
+      When I run command:
+        """
+        ${NVCF_CLI} --config ${REPO_ROOT}/tests/bdd/fixtures/nvcf-cli-local.yaml function invoke --inference-url /v1/chat/completions --model-name openai-compatible-sample --request-body '{"messages":[{"role":"user","content":"bdd-llm-echo"}]}' --timeout 120
+        """
+      Then the command exit code should be 0
+      And the command output should contain "chat.completion"
+      And the command output should contain "fixed 128-byte response"
+
+      # curl reports only the status code so the assertion cannot
+      # match response-body noise.
+      When I run command:
+        """
+        curl -s -o /dev/null -w "%{http_code}" -X POST http://llm.localhost:8080/v1/chat/completions -H "Content-Type: application/json" -d '{"model":"unauthenticated/check","messages":[]}'
+        """
+      Then the command exit code should be 0
+      And the command output should contain "401"
+
+      # Leave the GPU capacity free, same as the echo scenarios.
+      When I run command:
+        """
+        ${NVCF_CLI} --config ${REPO_ROOT}/tests/bdd/fixtures/nvcf-cli-local.yaml function delete --deployment-only
+        """
+      Then the command exit code should be 0
diff --git a/tests/bdd/fixtures/self-managed-local-bdd.yaml b/tests/bdd/fixtures/self-managed-local-bdd.yaml
index a746c6ed6..b4ed793ca 100644
--- a/tests/bdd/fixtures/self-managed-local-bdd.yaml
+++ b/tests/bdd/fixtures/self-managed-local-bdd.yaml
@@ -50,9 +50,13 @@ openbao:
 # Pylon exits at startup when the API advertises its default HTTPS dummy OTLP
 # endpoint because the current binary has no OpenTelemetry TLS provider. Keep
 # worker telemetry on HTTP in this fixture until Pylon supports HTTPS OTLP.
+#
+# The LLM router client sidecar image defaults to "dummy" in the
+# self-managed API profile; the Background overlays the placeholders.
 api:
   env:
     MANAGEMENT_OTLP_TRACING_ENDPOINT: http://dummy:8282
+    NVCF_SIDECARS_LLM_ROUTER_CLIENT_IMAGE: nvcr.io/REPLACE_WITH_SAMPLE_NGC_ORG/REPLACE_WITH_SAMPLE_NGC_TEAM/stargate-client:0.2.0
 
 # LLM gateway talks to nvcf-api over plaintext gRPC in local dev
 # (api.nvcf.svc.cluster.local:9090 has no TLS).
diff --git a/tests/bdd/godog_test.go b/tests/bdd/godog_test.go
index 91a8571f2..7b31c14cc 100644
--- a/tests/bdd/godog_test.go
+++ b/tests/bdd/godog_test.go
@@ -411,6 +411,18 @@ func TestSingleClusterHelmfileFeatureFileWiresToSteps(t *testing.T) {
 			ExitCode: 0,
 			Stdout:   "Function invocation completed!\n\nResponse:\n{\"message\":\"bdd-grpc-echo\"}\n",
 		},
+		"/usr/bin/nvcf-cli --config /repo-root-placeholder/tests/bdd/fixtures/nvcf-cli-local.yaml function invoke" +
+			" --inference-url /v1/chat/completions --model-name openai-compatible-sample" +
+			" --request-body '{\"messages\":[{\"role\":\"user\",\"content\":\"bdd-llm-echo\"}]}' --timeout 120": {
+			ExitCode: 0,
+			Stdout: "Function invocation completed!\n\nResponse:\n" +
+				`{"object":"chat.completion","choices":[{"message":{"content":"This is a fixed 128-byte response from an NVCF-hosted OpenAI-compatible sample, used for load testing and throughput benchmarks."}}]}` +
+				"\n",
+		},
+		`curl -s -o /dev/null -w "%{http_code}" -X POST http://llm.localhost:8080/v1/chat/completions -H "Content-Type: application/json" -d '{"model":"unauthenticated/check","messages":[]}'`: {
+			ExitCode: 0,
+			Stdout:   "401",
+		},
 		// Conflict precheck: feature asserts the conflicting
 		// multi-cluster control-plane is absent.
 		"k3d cluster get ncp-local-cp": {ExitCode: 1},
@@ -462,6 +474,22 @@ func TestSingleClusterHelmfileFeatureFileWiresToSteps(t *testing.T) {
 	if !commandRanThatContains(suite.Runner.(*fakeRunner).runs, "api-key generate --description bdd-grpc-load-tester-supreme --for function") {
 		t.Fatal("gRPC sample function API key was not generated for the function service")
 	}
+	if !commandRanThatContainsAll(suite.Runner.(*fakeRunner).runs,
+		"function create --name bdd-openai-compatible-sample",
+		"nvcf-openai-compatible-sample:local",
+		"--function-type LLM",
+		"--llm-model") {
+		t.Fatal("LLM sample function was not created with the LLM function type and model config")
+	}
+	if !commandRanThatContains(suite.Runner.(*fakeRunner).runs, "function invoke --inference-url /v1/chat/completions --model-name openai-compatible-sample") {
+		t.Fatal("LLM function invoke CLI command was never invoked")
+	}
+	if !commandRanThatContains(suite.Runner.(*fakeRunner).runs, "function delete --deployment-only") {
+		t.Fatal("function deployment cleanup was never invoked")
+	}
+	if !commandRanThatContains(suite.Runner.(*fakeRunner).runs, "http://llm.localhost:8080/v1/chat/completions") {
+		t.Fatal("unauthenticated LLM gateway check was never invoked")
+	}
 }
 
 // TestObservabilityControlFeatureFileWiresToSteps runs the live-install