diff --git a/.github/workflows/checks.yaml b/.github/workflows/checks.yaml
index 02946ff..a8e7329 100644
--- a/.github/workflows/checks.yaml
+++ b/.github/workflows/checks.yaml
@@ -30,7 +30,7 @@ jobs:
           conda activate cpp-conda
           mkdir build
           cd build
-          cmake ..
+          cmake -DCMAKE_PREFIX_PATH=`python -c 'import torch;print(torch.utils.cmake_prefix_path)'` ..
           make
 
       - name: Check Code Format
diff --git a/README.md b/README.md
index 12b94d1..24d8d1e 100644
--- a/README.md
+++ b/README.md
@@ -39,6 +39,8 @@ it _might_ work on other systems.
   - [xtensor](https://xtensor.readthedocs.io/en/latest/)
   - [nlohmann/json](https://github.com/nlohmann/json#serialization--deserialization)
   - [protobuf](https://github.com/protocolbuffers/protobuf)
+  - [PyTorch](https://pytorch.org) -> For CUDA support, you will need to install
+    the [CUDA toolkit](https://developer.nvidia.com/cuda-toolkit) yourself
 - Dev tools
   - Formatting (Google style) with
     [clang-format](https://clang.llvm.org/docs/ClangFormat.html)
@@ -77,7 +79,7 @@ conda activate cpp-conda
 ```bash
 mkdir build
 cd build
-cmake ..
+cmake -DCMAKE_PREFIX_PATH=`python -c 'import torch;print(torch.utils.cmake_prefix_path)'` ..
 make
 ```
 
diff --git a/environment.yaml b/environment.yaml
index 8b18ecf..77dc995 100644
--- a/environment.yaml
+++ b/environment.yaml
@@ -1,6 +1,7 @@
 name: cpp-conda
 channels:
   - conda-forge
+  - pytorch
   - defaults
 dependencies:
   - python=3.8.*
@@ -17,5 +18,6 @@ dependencies:
   - xsimd=7.*
   - nlohmann_json=3.*
   - protobuf=3.* # includes Protobuf compiler, C++ headers, Python libraries
+  - pytorch=1.7.*
   # Development
   - cmake_format
diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt
index f84a093..9627e92 100644
--- a/src/CMakeLists.txt
+++ b/src/CMakeLists.txt
@@ -3,6 +3,13 @@ find_package(xtensor REQUIRED)
 find_package(absl REQUIRED)
 find_package(nlohmann_json REQUIRED)
 
+#
+# PyTorch
+#
+
+find_package(Torch REQUIRED)
+set(CMAKE_CXX_FLAGS "${TORCH_CXX_FLAGS} ${CMAKE_CXX_FLAGS}")
+
 #
 # Protobuf
 #
@@ -23,8 +30,14 @@ message("PROTO_HDRS = ${PROTO_HDRS}")
 
 add_executable(main main.cpp ${PROTO_SRCS} ${PROTO_HDRS})
 target_link_libraries(
-  main PRIVATE xtensor xtensor::optimize xtensor::use_xsimd absl::flat_hash_map
-               private_library ${PROTOBUF_LIBRARIES})
+  main
+  PRIVATE xtensor
+          xtensor::optimize
+          xtensor::use_xsimd
+          absl::flat_hash_map
+          private_library
+          ${PROTOBUF_LIBRARIES}
+          ${TORCH_LIBRARIES})
 target_include_directories(
   main PRIVATE . ${PROTOBUF_INCLUDE_DIRS}
                ${CMAKE_CURRENT_BINARY_DIR} # In order to capture the *.pb.h
diff --git a/src/main.cpp b/src/main.cpp
index e3d9d5e..c55953b 100644
--- a/src/main.cpp
+++ b/src/main.cpp
@@ -2,6 +2,7 @@
 
 #include <absl/container/flat_hash_map.h>
 #include <nlohmann/json.hpp>
+#include <torch/torch.h>
 #include <xtensor/xarray.hpp>
 #include <xtensor/xio.hpp>
 #include <xtensor/xview.hpp>
@@ -50,5 +51,10 @@ int main() {
   std::cout << "Protobuf\n";
   projectname::AddressBook address_book;
 
+  std::cout << "Random torch tensor\n";
+  torch::Tensor tensor = torch::rand({2, 3});  // NOLINT
+  std::cout << tensor << std::endl;
+  std::cout << "cuda is available: " << torch::cuda::is_available() << '\n';
+
   return 0;
 }