diff --git a/pixi-cuda.toml b/pixi-cuda.toml
index 3dc03d1b6072712d3d7b8365c0a39ba3bf37080b..4c978eb200527f5115e646be44f90c24b4867f3c 100644
--- a/pixi-cuda.toml
+++ b/pixi-cuda.toml
@@ -1,3 +1,31 @@
+# SPDX-FileCopyrightText: Copyright © 2023 Idiap Research Institute <contact@idiap.ch>
+#
+# SPDX-License-Identifier: GPL-3.0-or-later
+
+# INTRODUCTION
+# ------------
+#
+# This pixi manifest file contains an example to create a CUDA-enabled version
+# of the mednet environment.  It loosely follows the recipe in the pixi
+# documentation: https://pixi.sh/latest/advanced/channel_priority/#use-case-pytorch-and-nvidia-with-conda-forge
+#
+# USAGE INSTRUCTIONS
+# ------------------
+#
+# To use this file, first link it to `pixi.toml`, and re-run `pixi install`.
+# Having a file named `pixi.toml` on the current directory has precedence over
+# `pyproject.toml`:
+#
+# $ ln -s pixi-cuda.toml pixi.toml
+# $ rm -rf .pixi pixi.lock  # clean-up
+# $ pixi install  # this will install all support libraries
+# $ pixi run build  # this may fail, see next
+#
+# If `pixi run build` does not work (in the case you are building on a machine
+# w/o CUDA support), then do the following:
+#
+# $ uv pip install --no-build-isolation --no-deps --editable ../mednet
+
 [project]
 name = "mednet"
 channels = ["nvidia", "conda-forge", "pytorch"]