diff --git a/k8s/clientpatch.go b/k8s/clientpatch.go
index b4ff770..26312ee 100644
--- a/k8s/clientpatch.go
+++ b/k8s/clientpatch.go
@@ -37,6 +37,14 @@ func PatchHibernateState(ctx context.Context, cli client.Client, pod *corev1.Pod
 	})
 }
 
+// PatchKeepSnapshotOnDelete flags the pod's deletion as a seat release, short-circuiting if already flagged.
+func PatchKeepSnapshotOnDelete(ctx context.Context, cli client.Client, pod *corev1.Pod) error {
+	if meta.ReadKeepSnapshotOnDelete(pod) {
+		return nil
+	}
+	return Patch(ctx, cli, pod, meta.MarkKeepSnapshotOnDelete)
+}
+
 // PatchCocoonSetGeneration stamps the owning CocoonSet's metadata.generation
 // onto the pod so vk-cocoon can read it back as lifecycle-observed-generation.
 // Short-circuits when the annotation is already correct.
diff --git a/k8s/clientpatch_test.go b/k8s/clientpatch_test.go
index 0d24ac0..1a33d67 100644
--- a/k8s/clientpatch_test.go
+++ b/k8s/clientpatch_test.go
@@ -113,6 +113,26 @@ func TestPatchCocoonSetGenerationShortCircuitsNoOp(t *testing.T) {
 	}
 }
 
+func TestPatchKeepSnapshotOnDeletePersistsAndShortCircuits(t *testing.T) {
+	pod := &corev1.Pod{ObjectMeta: metav1.ObjectMeta{Name: "demo", Namespace: "ns"}}
+	cli := newFakeClient(t, pod.DeepCopy())
+
+	if err := PatchKeepSnapshotOnDelete(t.Context(), cli, pod); err != nil {
+		t.Fatalf("PatchKeepSnapshotOnDelete: %v", err)
+	}
+	var got corev1.Pod
+	if err := cli.Get(t.Context(), client.ObjectKey{Namespace: "ns", Name: "demo"}, &got); err != nil {
+		t.Fatalf("get: %v", err)
+	}
+	if !meta.ReadKeepSnapshotOnDelete(&got) {
+		t.Errorf("flag must reach the API server before the delete lands: %v", got.Annotations)
+	}
+	// The fake client errors on an empty-body Patch, so success proves the no-op guard.
+	if err := PatchKeepSnapshotOnDelete(t.Context(), cli, &got); err != nil {
+		t.Fatalf("re-flagging an already-flagged pod must be a no-op: %v", err)
+	}
+}
+
 func newFakeClient(t *testing.T, objs ...client.Object) client.Client {
 	t.Helper()
 	scheme := runtime.NewScheme()
diff --git a/meta/hibernate.go b/meta/hibernate.go
index 79b8d19..aed15e9 100644
--- a/meta/hibernate.go
+++ b/meta/hibernate.go
@@ -34,3 +34,14 @@ func MarkRestoreFromHibernate(pod *corev1.Pod) {
 	a := ensurePodAnnotations(pod)
 	a[AnnotationRestoreFromHibernate] = annotationTrue
 }
+
+// ReadKeepSnapshotOnDelete reports whether the pod's deletion is flagged as a seat release.
+func ReadKeepSnapshotOnDelete(pod *corev1.Pod) bool {
+	return pod.Annotations[AnnotationKeepSnapshotOnDelete] == annotationTrue
+}
+
+// MarkKeepSnapshotOnDelete flags a pod's deletion as a seat release.
+func MarkKeepSnapshotOnDelete(pod *corev1.Pod) {
+	a := ensurePodAnnotations(pod)
+	a[AnnotationKeepSnapshotOnDelete] = annotationTrue
+}
diff --git a/meta/hibernate_test.go b/meta/hibernate_test.go
index b814fbd..fbf8e35 100644
--- a/meta/hibernate_test.go
+++ b/meta/hibernate_test.go
@@ -80,3 +80,14 @@ func TestDefaultSnapshotTagConstant(t *testing.T) {
 		t.Errorf("DefaultSnapshotTag must differ from HibernateSnapshotTag")
 	}
 }
+
+func TestMarkKeepSnapshotOnDelete(t *testing.T) {
+	pod := &corev1.Pod{}
+	if ReadKeepSnapshotOnDelete(pod) {
+		t.Fatal("fresh pod should not be flagged as a seat release")
+	}
+	MarkKeepSnapshotOnDelete(pod)
+	if !ReadKeepSnapshotOnDelete(pod) {
+		t.Error("MarkKeepSnapshotOnDelete should round-trip through ReadKeepSnapshotOnDelete")
+	}
+}
diff --git a/meta/keys.go b/meta/keys.go
index 51db498..aa18939 100644
--- a/meta/keys.go
+++ b/meta/keys.go
@@ -55,6 +55,8 @@ const (
 	// restore its VM from the :hibernate snapshot (cross-node migration) instead
 	// of cloning from the base image. Written by the operator on the rebuilt pod.
 	AnnotationRestoreFromHibernate = "vm.cocoonstack.io/restore-from-hibernate"
+	// AnnotationKeepSnapshotOnDelete requests node-local snapshot retention for a seat-release pod deletion.
+	AnnotationKeepSnapshotOnDelete = "vm.cocoonstack.io/keep-snapshot-on-delete"
 	// AnnotationForkFrom names a VM to fork the new VM from.
 	AnnotationForkFrom = "vm.cocoonstack.io/fork-from"
 	// AnnotationCloneFromDir names a host directory to clone the VM image from (vk-cocoon-specific).