diff --git a/ci/build_python.sh b/ci/build_python.sh
index 2e3f70ba767..823d7f62290 100755
--- a/ci/build_python.sh
+++ b/ci/build_python.sh
@@ -52,5 +52,10 @@ RAPIDS_PACKAGE_VERSION=$(head -1 ./VERSION) rapids-conda-retry mambabuild \
   --channel "${RAPIDS_CONDA_BLD_OUTPUT_DIR}" \
   conda/recipes/custreamz
 
+RAPIDS_PACKAGE_VERSION=$(head -1 ./VERSION) rapids-conda-retry mambabuild \
+  --no-test \
+  --channel "${CPP_CHANNEL}" \
+  --channel "${RAPIDS_CONDA_BLD_OUTPUT_DIR}" \
+  conda/recipes/cudf-polars
 
 rapids-upload-conda-to-s3 python
diff --git a/ci/test_python_other.sh b/ci/test_python_other.sh
index 21a59fa1494..db86721755d 100755
--- a/ci/test_python_other.sh
+++ b/ci/test_python_other.sh
@@ -14,7 +14,8 @@ rapids-mamba-retry install \
   --channel "${PYTHON_CHANNEL}" \
   "dask-cudf=${RAPIDS_VERSION}" \
   "cudf_kafka=${RAPIDS_VERSION}" \
-  "custreamz=${RAPIDS_VERSION}"
+  "custreamz=${RAPIDS_VERSION}" \
+  "cudf-polars=${RAPIDS_VERSION}"
 
 rapids-logger "Check GPU usage"
 nvidia-smi
@@ -37,7 +38,7 @@ rapids-logger "pytest dask_cudf (legacy)"
 DASK_DATAFRAME__QUERY_PLANNING=False ./ci/run_dask_cudf_pytests.sh \
   --junitxml="${RAPIDS_TESTS_DIR}/junit-dask-cudf-legacy.xml" \
   --numprocesses=8 \
-  --dist=loadscope \
+  --dist=worksteal \
   .
 
 rapids-logger "pytest cudf_kafka"
@@ -54,5 +55,19 @@ rapids-logger "pytest custreamz"
   --cov-report=xml:"${RAPIDS_COVERAGE_DIR}/custreamz-coverage.xml" \
   --cov-report=term
 
+# Note that cudf-polars uses rmm.mr.CudaAsyncMemoryResource() which allocates
+# half the available memory. This doesn't play well with multiple workers, so
+# we keep --numprocesses=1 for now. This should be resolved by
+# https://github.com/rapidsai/cudf/issues/16723.
+rapids-logger "pytest cudf-polars"
+./ci/run_cudf_polars_pytests.sh \
+  --junitxml="${RAPIDS_TESTS_DIR}/junit-cudf-polars.xml" \
+  --numprocesses=1 \
+  --dist=worksteal \
+  --cov-config=./pyproject.toml \
+  --cov=cudf_polars \
+  --cov-report=xml:"${RAPIDS_COVERAGE_DIR}/cudf-polars-coverage.xml" \
+  --cov-report=term
+
 rapids-logger "Test script exiting with value: $EXITCODE"
 exit ${EXITCODE}
diff --git a/conda/recipes/cudf-polars/build.sh b/conda/recipes/cudf-polars/build.sh
new file mode 100644
index 00000000000..06e2f1bcb99
--- /dev/null
+++ b/conda/recipes/cudf-polars/build.sh
@@ -0,0 +1,4 @@
+# Copyright (c) 2024, NVIDIA CORPORATION.
+
+# This assumes the script is executed from the root of the repo directory
+./build.sh cudf_polars
diff --git a/conda/recipes/cudf-polars/meta.yaml b/conda/recipes/cudf-polars/meta.yaml
new file mode 100644
index 00000000000..e8fef715c60
--- /dev/null
+++ b/conda/recipes/cudf-polars/meta.yaml
@@ -0,0 +1,61 @@
+# Copyright (c) 2024, NVIDIA CORPORATION.
+
+{% set version = environ['RAPIDS_PACKAGE_VERSION'].lstrip('v') %}
+{% set minor_version = version.split('.')[0] + '.' + version.split('.')[1] %}
+{% set py_version = environ['CONDA_PY'] %}
+{% set cuda_version = '.'.join(environ['RAPIDS_CUDA_VERSION'].split('.')[:2]) %}
+{% set cuda_major = cuda_version.split('.')[0] %}
+{% set date_string = environ['RAPIDS_DATE_STRING'] %}
+
+package:
+  name: cudf-polars
+  version: {{ version }}
+
+source:
+  path: ../../..
+
+build:
+  number: {{ GIT_DESCRIBE_NUMBER }}
+  string: cuda{{ cuda_major }}_py{{ py_version }}_{{ date_string }}_{{ GIT_DESCRIBE_HASH }}_{{ GIT_DESCRIBE_NUMBER }}
+  script_env:
+    - AWS_ACCESS_KEY_ID
+    - AWS_SECRET_ACCESS_KEY
+    - AWS_SESSION_TOKEN
+    - CMAKE_C_COMPILER_LAUNCHER
+    - CMAKE_CUDA_COMPILER_LAUNCHER
+    - CMAKE_CXX_COMPILER_LAUNCHER
+    - CMAKE_GENERATOR
+    - PARALLEL_LEVEL
+    - SCCACHE_BUCKET
+    - SCCACHE_IDLE_TIMEOUT
+    - SCCACHE_REGION
+    - SCCACHE_S3_KEY_PREFIX=cudf-polars-aarch64 # [aarch64]
+    - SCCACHE_S3_KEY_PREFIX=cudf-polars-linux64 # [linux64]
+    - SCCACHE_S3_USE_SSL
+    - SCCACHE_S3_NO_CREDENTIALS
+
+requirements:
+  host:
+    - python
+    - rapids-build-backend >=0.3.0,<0.4.0.dev0
+    - setuptools
+    - cuda-version ={{ cuda_version }}
+  run:
+    - python
+    - pylibcudf ={{ version }}
+    - polars >=1.8,<1.9
+    - {{ pin_compatible('cuda-version', max_pin='x', min_pin='x') }}
+
+test:
+  requires:
+    - cuda-version ={{ cuda_version }}
+  imports:
+    - cudf_polars
+
+
+about:
+  home: https://rapids.ai/
+  license: Apache-2.0
+  license_family: APACHE
+  license_file: LICENSE
+  summary: cudf-polars library
diff --git a/python/cudf_polars/tests/expressions/test_agg.py b/python/cudf_polars/tests/expressions/test_agg.py
index 3001a61101a..86cb2352dcc 100644
--- a/python/cudf_polars/tests/expressions/test_agg.py
+++ b/python/cudf_polars/tests/expressions/test_agg.py
@@ -96,7 +96,7 @@ def test_bool_agg(agg, request):
     assert_gpu_result_equal(q, check_exact=False)
 
 
-@pytest.mark.parametrize("cum_agg", expr.UnaryFunction._supported_cum_aggs)
+@pytest.mark.parametrize("cum_agg", sorted(expr.UnaryFunction._supported_cum_aggs))
 def test_cum_agg_reverse_unsupported(cum_agg):
     df = pl.LazyFrame({"a": [1, 2, 3]})
     expr = getattr(pl.col("a"), cum_agg)(reverse=True)