@@ -38,6 +38,9 @@ Kernels and launch configuration
 - :cuda-core-example:`host_launch.py <host_launch.py>`
   orders a host function between two kernel launches on a stream and reads a
   pinned buffer from it.
+- :cuda-core-example:`host_copy.py <host_copy.py>`
+  copies a device buffer into a NumPy array in chunks through two pinned
+  staging slots, with the host copies done by ``HOST_COPY_FN``.
 
 Memory management and interop
 -----------------------------
 
@@ -0,0 +1,61 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+#
+# SPDX-License-Identifier: Apache-2.0
+
+# ################################################################################
+#
+# This example copies a device buffer into a NumPy array in chunks through two
+# pinned staging slots. Each chunk is a device-to-pinned copy followed by
+# HOST_COPY_FN, which copies the slot into the array without the GIL. Each slot
+# has its own stream, so the copy into one slot overlaps the host copy out of
+# the other, and stream order keeps a slot from being reused too early.
+#
+# ################################################################################
+
+# /// script
+# dependencies = ["cuda_bindings", "cuda_core", "numpy"]
+# ///
+
+from contextlib import closing
+
+import numpy as np
+
+from cuda.core import HOST_COPY_FN, Buffer, Device, LegacyPinnedMemoryResource, host_launch, pack_host_copy_args
+
+
+def main():
+    dev = Device(0)
+    dev.set_current()
+    chunk = 4 << 20
+    total = 16 * chunk
+    result = np.empty(total, dtype=np.uint8)
+    pinned_mr = LegacyPinnedMemoryResource()
+
+    with (
+        closing(dev.create_stream()) as stream0,
+        closing(dev.create_stream()) as stream1,
+        dev.allocate(total, stream=stream0) as device_buffer,
+        pinned_mr.allocate(chunk) as slot0,
+        pinned_mr.allocate(chunk) as slot1,
+    ):
+        device_buffer.fill(0x5A, stream=stream0)
+        stream1.wait(stream0)
+
+        streams = (stream0, stream1)
+        slots = (slot0, slot1)
+        for i, offset in enumerate(range(0, total, chunk)):
+            stream, slot = streams[i % 2], slots[i % 2]
+            src = Buffer.from_handle(int(device_buffer.handle) + offset, chunk, owner=device_buffer)
+            src.copy_to(slot, stream=stream)
+            dst = memoryview(result)[offset : offset + chunk]
+            host_launch(stream, HOST_COPY_FN, user_data=pack_host_copy_args(dst, slot, chunk))
+
+        stream0.sync()
+        stream1.sync()
+
+    assert (result == 0x5A).all()
+    print(f"Copied {total >> 20} MiB into a NumPy array through two pinned slots.")
+
+
+if __name__ == "__main__":
+    main()