diff --git a/.devcontainer/README.md b/.devcontainer/README.md
index cc5fac22fde2..91ee7ef85f78 100644
--- a/.devcontainer/README.md
+++ b/.devcontainer/README.md
@@ -20,7 +20,6 @@ This container is a turnkey development environment for building and testing the
 By default, the following directories are bind-mounted into the devcontainer:
 
 * `${repo}:/home/coder/cudf`
-* `${repo}/../rapidsmpf:/home/coder/rapidsmpf`
 * `${repo}/../.aws:/home/coder/.aws`
 * `${repo}/../.local:/home/coder/.local`
 * `${repo}/../.cache:/home/coder/.cache`
@@ -29,13 +28,6 @@ By default, the following directories are bind-mounted into the devcontainer:
 
 This ensures caches, configurations, dependencies, and your commits are persisted on the host across container runs.
 
-The [rapidsmpf](https://github.com/rapidsai/rapidsmpf) repository is a required dependency of `cudf_polars` (that also requires `libcudf`) and must be cloned as a sibling directory to the cudf repo before launching the devcontainer:
-
-```
-# from the parent directory of your cudf clone
-git clone https://github.com/rapidsai/rapidsmpf.git
-```
-
 ## Launch a Dev Container
 
 To launch a devcontainer from VSCode, open the cuDF repo and select the "Reopen in Container" button in the bottom right:<br/><img src="https://user-images.githubusercontent.com/178183/221771999-97ab29d5-e718-4e5f-b32f-2cdd51bba25c.png"/>
diff --git a/.devcontainer/cuda12.9-conda/devcontainer.json b/.devcontainer/cuda12.9-conda/devcontainer.json
index f8581936a262..5326e64d8c6c 100644
--- a/.devcontainer/cuda12.9-conda/devcontainer.json
+++ b/.devcontainer/cuda12.9-conda/devcontainer.json
@@ -52,7 +52,7 @@
   "initializeCommand": [
     "/bin/bash",
     "-c",
-    "mkdir -m 0755 -p ${localWorkspaceFolder}/../.{aws,cache,config,conda/pkgs,conda/${localWorkspaceFolderBasename}-cuda12.9-envs} ${localWorkspaceFolder}/../rapidsmpf"
+    "mkdir -m 0755 -p ${localWorkspaceFolder}/../.{aws,cache,config,conda/pkgs,conda/${localWorkspaceFolderBasename}-cuda12.9-envs}"
   ],
   "postAttachCommand": [
     "/bin/bash",
@@ -66,8 +66,7 @@
     "source=${localWorkspaceFolder}/../.cache,target=/home/coder/.cache,type=bind,consistency=consistent",
     "source=${localWorkspaceFolder}/../.config,target=/home/coder/.config,type=bind,consistency=consistent",
     "source=${localWorkspaceFolder}/../.conda/pkgs,target=/home/coder/.conda/pkgs,type=bind,consistency=consistent",
-    "source=${localWorkspaceFolder}/../.conda/${localWorkspaceFolderBasename}-cuda12.9-envs,target=/home/coder/.conda/envs,type=bind,consistency=consistent",
-    "source=${localWorkspaceFolder}/../rapidsmpf,target=/home/coder/rapidsmpf,type=bind,consistency=consistent"
+    "source=${localWorkspaceFolder}/../.conda/${localWorkspaceFolderBasename}-cuda12.9-envs,target=/home/coder/.conda/envs,type=bind,consistency=consistent"
   ],
   "customizations": {
     "vscode": {
diff --git a/.devcontainer/cuda12.9-pip/devcontainer.json b/.devcontainer/cuda12.9-pip/devcontainer.json
index 03851a926c8c..914237875c0f 100644
--- a/.devcontainer/cuda12.9-pip/devcontainer.json
+++ b/.devcontainer/cuda12.9-pip/devcontainer.json
@@ -30,7 +30,7 @@
   "initializeCommand": [
     "/bin/bash",
     "-c",
-    "mkdir -m 0755 -p ${localWorkspaceFolder}/../.{aws,cache,config/pip,local/share/${localWorkspaceFolderBasename}-cuda12.9-venvs} ${localWorkspaceFolder}/../rapidsmpf"
+    "mkdir -m 0755 -p ${localWorkspaceFolder}/../.{aws,cache,config/pip,local/share/${localWorkspaceFolderBasename}-cuda12.9-venvs}"
   ],
   "postAttachCommand": [
     "/bin/bash",
@@ -43,8 +43,7 @@
     "source=${localWorkspaceFolder}/../.aws,target=/home/coder/.aws,type=bind,consistency=consistent",
     "source=${localWorkspaceFolder}/../.cache,target=/home/coder/.cache,type=bind,consistency=consistent",
     "source=${localWorkspaceFolder}/../.config,target=/home/coder/.config,type=bind,consistency=consistent",
-    "source=${localWorkspaceFolder}/../.local/share/${localWorkspaceFolderBasename}-cuda12.9-venvs,target=/home/coder/.local/share/venvs,type=bind,consistency=consistent",
-    "source=${localWorkspaceFolder}/../rapidsmpf,target=/home/coder/rapidsmpf,type=bind,consistency=consistent"
+    "source=${localWorkspaceFolder}/../.local/share/${localWorkspaceFolderBasename}-cuda12.9-venvs,target=/home/coder/.local/share/venvs,type=bind,consistency=consistent"
   ],
   "customizations": {
     "vscode": {
diff --git a/.devcontainer/cuda13.2-conda/devcontainer.json b/.devcontainer/cuda13.2-conda/devcontainer.json
index 78a2f7f64963..a7987bd05dff 100644
--- a/.devcontainer/cuda13.2-conda/devcontainer.json
+++ b/.devcontainer/cuda13.2-conda/devcontainer.json
@@ -52,7 +52,7 @@
   "initializeCommand": [
     "/bin/bash",
     "-c",
-    "mkdir -m 0755 -p ${localWorkspaceFolder}/../.{aws,cache,config,conda/pkgs,conda/${localWorkspaceFolderBasename}-cuda13.2-envs} ${localWorkspaceFolder}/../rapidsmpf"
+    "mkdir -m 0755 -p ${localWorkspaceFolder}/../.{aws,cache,config,conda/pkgs,conda/${localWorkspaceFolderBasename}-cuda13.2-envs}"
   ],
   "postAttachCommand": [
     "/bin/bash",
@@ -66,8 +66,7 @@
     "source=${localWorkspaceFolder}/../.cache,target=/home/coder/.cache,type=bind,consistency=consistent",
     "source=${localWorkspaceFolder}/../.config,target=/home/coder/.config,type=bind,consistency=consistent",
     "source=${localWorkspaceFolder}/../.conda/pkgs,target=/home/coder/.conda/pkgs,type=bind,consistency=consistent",
-    "source=${localWorkspaceFolder}/../.conda/${localWorkspaceFolderBasename}-cuda13.2-envs,target=/home/coder/.conda/envs,type=bind,consistency=consistent",
-    "source=${localWorkspaceFolder}/../rapidsmpf,target=/home/coder/rapidsmpf,type=bind,consistency=consistent"
+    "source=${localWorkspaceFolder}/../.conda/${localWorkspaceFolderBasename}-cuda13.2-envs,target=/home/coder/.conda/envs,type=bind,consistency=consistent"
   ],
   "customizations": {
     "vscode": {
diff --git a/.devcontainer/cuda13.2-pip/devcontainer.json b/.devcontainer/cuda13.2-pip/devcontainer.json
index 7a283715b2e8..181ba40473cc 100644
--- a/.devcontainer/cuda13.2-pip/devcontainer.json
+++ b/.devcontainer/cuda13.2-pip/devcontainer.json
@@ -30,7 +30,7 @@
   "initializeCommand": [
     "/bin/bash",
     "-c",
-    "mkdir -m 0755 -p ${localWorkspaceFolder}/../.{aws,cache,config/pip,local/share/${localWorkspaceFolderBasename}-cuda13.2-venvs} ${localWorkspaceFolder}/../rapidsmpf"
+    "mkdir -m 0755 -p ${localWorkspaceFolder}/../.{aws,cache,config/pip,local/share/${localWorkspaceFolderBasename}-cuda13.2-venvs}"
   ],
   "postAttachCommand": [
     "/bin/bash",
@@ -43,8 +43,7 @@
     "source=${localWorkspaceFolder}/../.aws,target=/home/coder/.aws,type=bind,consistency=consistent",
     "source=${localWorkspaceFolder}/../.cache,target=/home/coder/.cache,type=bind,consistency=consistent",
     "source=${localWorkspaceFolder}/../.config,target=/home/coder/.config,type=bind,consistency=consistent",
-    "source=${localWorkspaceFolder}/../.local/share/${localWorkspaceFolderBasename}-cuda13.2-venvs,target=/home/coder/.local/share/venvs,type=bind,consistency=consistent",
-    "source=${localWorkspaceFolder}/../rapidsmpf,target=/home/coder/rapidsmpf,type=bind,consistency=consistent"
+    "source=${localWorkspaceFolder}/../.local/share/${localWorkspaceFolderBasename}-cuda13.2-venvs,target=/home/coder/.local/share/venvs,type=bind,consistency=consistent"
   ],
   "customizations": {
     "vscode": {
diff --git a/.github/workflows/pr.yaml b/.github/workflows/pr.yaml
index 031e26e4a048..1a345e4a498f 100644
--- a/.github/workflows/pr.yaml
+++ b/.github/workflows/pr.yaml
@@ -646,7 +646,7 @@ jobs:
       package-type: python
       pure-wheel: true
   wheel-tests-cudf-polars:
-    needs: [wheel-build-cudf-polars, changed-files]
+    needs: [wheel-build-cudf-polars, wheel-build-cudf-streaming, changed-files]
     permissions:
       actions: read
       contents: read
@@ -663,7 +663,7 @@ jobs:
       container-options: "--cap-add CAP_SYS_PTRACE --shm-size=8g --ulimit=nofile=1000000:1000000"
       script: "ci/test_wheel_cudf_polars.sh"
   cudf-polars-polars-tests:
-    needs: [wheel-build-cudf-polars, changed-files]
+    needs: [wheel-build-cudf-polars, wheel-build-cudf-streaming, changed-files]
     permissions:
       actions: read
       contents: read
@@ -732,14 +732,9 @@ jobs:
         SCCACHE_DIST_MAX_RETRIES=inf
         SCCACHE_SERVER_LOG=sccache=debug
         SCCACHE_DIST_FALLBACK_TO_LOCAL_COMPILE=false
-      # Call clone-rapidsmpf before creating the Python env to exclude cudf_polars' dependency on rapidsmpf
-      # librapidsmpf-cu13 wheels brings in a hardcoded libnuma-dev cmake target: https://github.com/NVIDIA/cuCascade/issues/118
-      # -DBUILD_TESTS=OFF to match rapidsmpf https://github.com/rapidsai/rapidsmpf/blob/main/.github/workflows/pr.yaml#L351 (leads to compilation errors)
-      # -DCUDF_BUILD_TESTUTIL=OFF to avoid IMPORTED_GLOBAL promotion errors when cuCascade's find_package(cudf) loads cudf-config.cmake from a CPM subdirectory
       build_command: |
         sccache --zero-stats;
-        clone-rapidsmpf -j$(nproc) -v -q --branch "$(cat ~/cudf/RAPIDS_BRANCH)" --clone-upstream --depth 1 --single-branch --shallow-submodules;
-        build-all -j0 -DBUILD_BENCHMARKS=OFF -DBUILD_NUMA_SUPPORT=OFF -DBUILD_TESTS=OFF -DCUDF_BUILD_TESTUTIL=OFF --verbose 2>&1 | tee telemetry-artifacts/build.log;
+        build-all -j0 -DBUILD_BENCHMARKS=ON --verbose 2>&1 | tee telemetry-artifacts/build.log;
         sccache --show-adv-stats | tee telemetry-artifacts/sccache-stats.txt;
   unit-tests-cudf-pandas:
     needs: [wheel-build-cudf, changed-files]
diff --git a/ci/test_cudf_polars_polars_tests.sh b/ci/test_cudf_polars_polars_tests.sh
index 1fb9a4e395f6..50d1ee861f16 100755
--- a/ci/test_cudf_polars_polars_tests.sh
+++ b/ci/test_cudf_polars_polars_tests.sh
@@ -12,6 +12,8 @@ RAPIDS_PY_CUDA_SUFFIX="$(rapids-wheel-ctk-name-gen "${RAPIDS_CUDA_VERSION}")"
 CUDF_POLARS_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="cudf_polars_${RAPIDS_PY_CUDA_SUFFIX}" RAPIDS_PY_WHEEL_PURE="1" rapids-download-wheels-from-github python)
 LIBCUDF_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="libcudf_${RAPIDS_PY_CUDA_SUFFIX}" rapids-download-wheels-from-github cpp)
 PYLIBCUDF_WHEELHOUSE=$(rapids-download-from-github "$(rapids-package-name "wheel_python" pylibcudf --stable --cuda "$RAPIDS_CUDA_VERSION")")
+LIBCUDF_STREAMING_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="libcudf_streaming_${RAPIDS_PY_CUDA_SUFFIX}" rapids-download-wheels-from-github cpp)
+CUDF_STREAMING_WHEELHOUSE=$(rapids-download-from-github "$(rapids-package-name "wheel_python" cudf_streaming --stable --cuda "$RAPIDS_CUDA_VERSION")")
 
 # generate constraints (possibly pinning to oldest support versions of dependencies)
 rapids-generate-pip-constraints py_test_cudf_polars "${PIP_CONSTRAINT}"
@@ -29,7 +31,9 @@ rapids-pip-retry install \
     --constraint "${PIP_CONSTRAINT}" \
     "$(echo "${CUDF_POLARS_WHEELHOUSE}"/cudf_polars_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)[test]" \
     "$(echo "${LIBCUDF_WHEELHOUSE}"/libcudf_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)" \
-    "$(echo "${PYLIBCUDF_WHEELHOUSE}"/pylibcudf_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)"
+    "$(echo "${PYLIBCUDF_WHEELHOUSE}"/pylibcudf_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)" \
+    "$(echo "${LIBCUDF_STREAMING_WHEELHOUSE}"/libcudf_streaming_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)" \
+    "$(echo "${CUDF_STREAMING_WHEELHOUSE}"/cudf_streaming_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)"
 
 TAG=$(python -c 'import polars; print(f"py-{polars.__version__}")')
 rapids-logger "Clone polars to ${TAG}"
diff --git a/ci/test_wheel_cudf_polars.sh b/ci/test_wheel_cudf_polars.sh
index 0b73e904dc47..61185c21568a 100755
--- a/ci/test_wheel_cudf_polars.sh
+++ b/ci/test_wheel_cudf_polars.sh
@@ -15,6 +15,10 @@ CUDF_POLARS_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="cudf_polars_${RAPIDS_PY_CUDA_SUFF
 LIBCUDF_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="libcudf_${RAPIDS_PY_CUDA_SUFFIX}" rapids-download-wheels-from-github cpp)
 PYLIBCUDF_WHEELHOUSE=$(rapids-download-from-github "$(rapids-package-name "wheel_python" pylibcudf --stable --cuda "$RAPIDS_CUDA_VERSION")")
 
+# Download libcudf_streaming and cudf_streaming built in the previous step
+LIBCUDF_STREAMING_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="libcudf_streaming_${RAPIDS_PY_CUDA_SUFFIX}" rapids-download-wheels-from-github cpp)
+CUDF_STREAMING_WHEELHOUSE=$(rapids-download-from-github "$(rapids-package-name "wheel_python" cudf_streaming --stable --cuda "$RAPIDS_CUDA_VERSION")")
+
 # generate constraints (possibly pinning to oldest support versions of dependencies)
 rapids-generate-pip-constraints py_test_cudf_polars "${PIP_CONSTRAINT}"
 
@@ -65,6 +69,8 @@ for version in "${VERSIONS[@]}"; do
         "$(echo "${CUDF_POLARS_WHEELHOUSE}"/cudf_polars_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)[test,dask,ray]" \
         "$(echo "${LIBCUDF_WHEELHOUSE}"/libcudf_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)" \
         "$(echo "${PYLIBCUDF_WHEELHOUSE}"/pylibcudf_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)" \
+        "$(echo "${LIBCUDF_STREAMING_WHEELHOUSE}"/libcudf_streaming_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)" \
+        "$(echo "${CUDF_STREAMING_WHEELHOUSE}"/cudf_streaming_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)" \
         -r "polars-compat-${version}-requirements.txt"
 
     rapids-logger "Running tests for polars ${version}.*"
diff --git a/conda/environments/all_cuda-129_arch-aarch64.yaml b/conda/environments/all_cuda-129_arch-aarch64.yaml
index 838a2c557b47..7e0ad88ddb96 100644
--- a/conda/environments/all_cuda-129_arch-aarch64.yaml
+++ b/conda/environments/all_cuda-129_arch-aarch64.yaml
@@ -44,6 +44,7 @@ dependencies:
 - libkvikio==26.8.*,>=0.0.0a0
 - libnvcomp-dev==5.2.0.10
 - libnvjitlink-dev
+- librapidsmpf==26.8.*,>=0.0.0a0
 - librdkafka>=2.11.1,<2.12.0
 - librmm==26.8.*,>=0.0.0a0
 - make
@@ -87,6 +88,7 @@ dependencies:
 - rapids-build-backend>=0.4.0,<0.5.0
 - rapids-dask-dependency==26.8.*,>=0.0.0a0
 - rapids-logger==0.2.*,>=0.0.0a0
+- rapidsmpf==26.8.*,>=0.0.0a0
 - ray-default>=2.55.1
 - rich
 - rmm==26.8.*,>=0.0.0a0
diff --git a/conda/environments/all_cuda-129_arch-x86_64.yaml b/conda/environments/all_cuda-129_arch-x86_64.yaml
index 0a435504e32f..272914ad7635 100644
--- a/conda/environments/all_cuda-129_arch-x86_64.yaml
+++ b/conda/environments/all_cuda-129_arch-x86_64.yaml
@@ -44,6 +44,7 @@ dependencies:
 - libkvikio==26.8.*,>=0.0.0a0
 - libnvcomp-dev==5.2.0.10
 - libnvjitlink-dev
+- librapidsmpf==26.8.*,>=0.0.0a0
 - librdkafka>=2.11.1,<2.12.0
 - librmm==26.8.*,>=0.0.0a0
 - make
@@ -87,6 +88,7 @@ dependencies:
 - rapids-build-backend>=0.4.0,<0.5.0
 - rapids-dask-dependency==26.8.*,>=0.0.0a0
 - rapids-logger==0.2.*,>=0.0.0a0
+- rapidsmpf==26.8.*,>=0.0.0a0
 - ray-default>=2.55.1
 - rich
 - rmm==26.8.*,>=0.0.0a0
diff --git a/conda/environments/all_cuda-132_arch-aarch64.yaml b/conda/environments/all_cuda-132_arch-aarch64.yaml
index 2093888e01fa..9ecba408142c 100644
--- a/conda/environments/all_cuda-132_arch-aarch64.yaml
+++ b/conda/environments/all_cuda-132_arch-aarch64.yaml
@@ -44,6 +44,7 @@ dependencies:
 - libkvikio==26.8.*,>=0.0.0a0
 - libnvcomp-dev==5.2.0.10
 - libnvjitlink-dev
+- librapidsmpf==26.8.*,>=0.0.0a0
 - librdkafka>=2.11.1,<2.12.0
 - librmm==26.8.*,>=0.0.0a0
 - make
@@ -87,6 +88,7 @@ dependencies:
 - rapids-build-backend>=0.4.0,<0.5.0
 - rapids-dask-dependency==26.8.*,>=0.0.0a0
 - rapids-logger==0.2.*,>=0.0.0a0
+- rapidsmpf==26.8.*,>=0.0.0a0
 - ray-default>=2.55.1
 - rich
 - rmm==26.8.*,>=0.0.0a0
diff --git a/conda/environments/all_cuda-132_arch-x86_64.yaml b/conda/environments/all_cuda-132_arch-x86_64.yaml
index 360550447330..ced21db00730 100644
--- a/conda/environments/all_cuda-132_arch-x86_64.yaml
+++ b/conda/environments/all_cuda-132_arch-x86_64.yaml
@@ -44,6 +44,7 @@ dependencies:
 - libkvikio==26.8.*,>=0.0.0a0
 - libnvcomp-dev==5.2.0.10
 - libnvjitlink-dev
+- librapidsmpf==26.8.*,>=0.0.0a0
 - librdkafka>=2.11.1,<2.12.0
 - librmm==26.8.*,>=0.0.0a0
 - make
@@ -87,6 +88,7 @@ dependencies:
 - rapids-build-backend>=0.4.0,<0.5.0
 - rapids-dask-dependency==26.8.*,>=0.0.0a0
 - rapids-logger==0.2.*,>=0.0.0a0
+- rapidsmpf==26.8.*,>=0.0.0a0
 - ray-default>=2.55.1
 - rich
 - rmm==26.8.*,>=0.0.0a0
diff --git a/conda/recipes/cudf-polars/recipe.yaml b/conda/recipes/cudf-polars/recipe.yaml
index 79ebec2963cb..12aa1e19f162 100644
--- a/conda/recipes/cudf-polars/recipe.yaml
+++ b/conda/recipes/cudf-polars/recipe.yaml
@@ -37,6 +37,7 @@ requirements:
     - python
     - pylibcudf =${{ version }}
     - rapidsmpf =${{ minor_version }}
+    - cudf-streaming =${{ version }}
     - polars>=1.35,<1.42
     - packaging
     - ${{ pin_compatible("cuda-version", upper_bound="x", lower_bound="x") }}
diff --git a/cpp/CMakeLists.txt b/cpp/CMakeLists.txt
index 064549c401dd..960c1c61eb43 100644
--- a/cpp/CMakeLists.txt
+++ b/cpp/CMakeLists.txt
@@ -290,6 +290,16 @@ endfunction()
 # * conda environment -----------------------------------------------------------------------------
 rapids_cmake_support_conda_env(conda_env MODIFY_PREFIX_PATH)
 
+# ##################################################################################################
+# * find C++ libraries from RAPIDS wheels in in site-packages --------------------------------------
+find_package(Python COMPONENTS Interpreter)
+if(Python_FOUND)
+  include(${rapids-cmake-dir}/cython-core/find_prefix_paths.cmake)
+  rapids_cython_find_prefix_paths("${Python_EXECUTABLE}" wheel_prefix_paths)
+  set_property(GLOBAL PROPERTY FIND_LIBRARY_USE_LIB64_PATHS TRUE)
+  list(APPEND CMAKE_PREFIX_PATH ${wheel_prefix_paths})
+endif()
+
 # ##################################################################################################
 # * compiler options ------------------------------------------------------------------------------
 set(CMAKE_MODULE_PATH ${CMAKE_CURRENT_SOURCE_DIR}/cmake/Modules ${CMAKE_MODULE_PATH})
diff --git a/dependencies.yaml b/dependencies.yaml
index 25ed1071f03d..0409d0c4cfe2 100644
--- a/dependencies.yaml
+++ b/dependencies.yaml
@@ -23,8 +23,10 @@ files:
       - depends_on_libkvikio
       - depends_on_librmm
       - depends_on_libnvcomp
+      - depends_on_librapidsmpf
       - depends_on_numba_cuda
       - depends_on_rapids_logger
+      - depends_on_rapidsmpf
       - depends_on_ray
       - depends_on_rmm
       - develop
@@ -335,6 +337,7 @@ files:
     includes:
       - run_cudf_polars
       - depends_on_rapidsmpf
+      - depends_on_cudf_streaming
       - depends_on_pylibcudf
       - depends_on_cuda_python
   # TODO(26.10): Remove this alias in favor of dask
@@ -1414,7 +1417,27 @@ dependencies:
     common:
       - output_types: conda
         packages:
-          - cudf-streaming==26.8.*,>=0.0.0a0
+          - &cudf_streaming_unsuffixed cudf-streaming==26.8.*,>=0.0.0a0
+      - output_types: requirements
+        packages:
+          # pip recognizes the index as a global option for the requirements.txt file
+          - --extra-index-url=https://pypi.anaconda.org/rapidsai-wheels-nightly/simple
+    specific:
+      - output_types: [requirements, pyproject]
+        matrices:
+          - matrix:
+              cuda: "12.*"
+              cuda_suffixed: "true"
+            packages:
+              - cudf-streaming-cu12==26.8.*,>=0.0.0a0
+          - matrix:
+              cuda: "13.*"
+              cuda_suffixed: "true"
+            packages:
+              - cudf-streaming-cu13==26.8.*,>=0.0.0a0
+          - matrix:
+            packages:
+              - *cudf_streaming_unsuffixed
   depends_on_ray:
     common:
       - output_types: conda
diff --git a/python/cudf_polars/cudf_polars/engine/core.py b/python/cudf_polars/cudf_polars/engine/core.py
index 79914d12fe8d..bfe97050ccf3 100644
--- a/python/cudf_polars/cudf_polars/engine/core.py
+++ b/python/cudf_polars/cudf_polars/engine/core.py
@@ -14,14 +14,15 @@
 from typing import TYPE_CHECKING, Any, ClassVar, Self, TypeVar
 
 import cuda.core
+
+import polars as pl
+
+from cudf_streaming.streaming.table_chunk import TableChunk
 from rapidsmpf.coll import AllGather
 from rapidsmpf.config import Options, get_environment_variables
 from rapidsmpf.memory.packed_data import PackedData
 from rapidsmpf.statistics import Statistics
 from rapidsmpf.streaming.core.actor import run_actor_network
-from rapidsmpf.streaming.cudf.table_chunk import TableChunk
-
-import polars as pl
 
 from cudf_polars.containers import DataFrame
 from cudf_polars.dsl.ir import IRExecutionContext
@@ -42,10 +43,10 @@
     from concurrent.futures import Executor, ThreadPoolExecutor
 
     import rapidsmpf.config
+    from cudf_streaming.streaming.channel_metadata import ChannelMetadata
     from rapidsmpf.communicator.communicator import Communicator
     from rapidsmpf.memory.buffer_resource import BufferResource
     from rapidsmpf.streaming.core.context import Context
-    from rapidsmpf.streaming.cudf.channel_metadata import ChannelMetadata
 
     from cudf_polars.dsl.ir import IR
     from cudf_polars.streaming.base import PartitionInfo
diff --git a/python/cudf_polars/cudf_polars/engine/dask.py b/python/cudf_polars/cudf_polars/engine/dask.py
index 014627b82c48..46372d0dad04 100644
--- a/python/cudf_polars/cudf_polars/engine/dask.py
+++ b/python/cudf_polars/cudf_polars/engine/dask.py
@@ -16,6 +16,10 @@
 import distributed.system
 import pynvml
 import ucxx._lib.libucxx as ucx_api
+
+import polars as pl
+
+import rmm.mr
 from rapidsmpf import bootstrap
 from rapidsmpf.communicator.ucxx import barrier, get_root_ucxx_address, new_communicator
 from rapidsmpf.config import Options
@@ -24,10 +28,6 @@
 from rapidsmpf.statistics import Statistics
 from rapidsmpf.streaming.core.context import Context
 
-import polars as pl
-
-import rmm.mr
-
 from cudf_polars.engine.core import (
     ClusterInfo,
     StreamingEngine,
@@ -44,8 +44,8 @@
 if TYPE_CHECKING:
     from collections.abc import Callable
 
+    from cudf_streaming.streaming.channel_metadata import ChannelMetadata
     from rapidsmpf.communicator.communicator import Communicator
-    from rapidsmpf.streaming.cudf.channel_metadata import ChannelMetadata
 
     from cudf_polars.dsl.ir import IR
     from cudf_polars.engine.core import T
diff --git a/python/cudf_polars/cudf_polars/engine/ray.py b/python/cudf_polars/cudf_polars/engine/ray.py
index edf1905e916d..b8824ee55414 100644
--- a/python/cudf_polars/cudf_polars/engine/ray.py
+++ b/python/cudf_polars/cudf_polars/engine/ray.py
@@ -12,6 +12,10 @@
 
 import ray
 import ucxx._lib.libucxx as ucx_api
+
+import polars as pl
+
+import rmm.mr
 from rapidsmpf import bootstrap
 from rapidsmpf.communicator.ucxx import barrier, get_root_ucxx_address, new_communicator
 from rapidsmpf.config import Options
@@ -20,10 +24,6 @@
 from rapidsmpf.statistics import Statistics
 from rapidsmpf.streaming.core.context import Context
 
-import polars as pl
-
-import rmm.mr
-
 from cudf_polars.engine.core import (
     ClusterInfo,
     StreamingEngine,
@@ -41,10 +41,11 @@
     import uuid
     from collections.abc import Callable
 
-    from rapidsmpf.communicator.communicator import Communicator
-    from rapidsmpf.streaming.cudf.channel_metadata import ChannelMetadata
     from ray.actor import ActorHandle
 
+    from cudf_streaming.streaming.channel_metadata import ChannelMetadata
+    from rapidsmpf.communicator.communicator import Communicator
+
     from cudf_polars.dsl.ir import IR
     from cudf_polars.engine.core import T
     from cudf_polars.engine.options import StreamingOptions
diff --git a/python/cudf_polars/cudf_polars/engine/spmd.py b/python/cudf_polars/cudf_polars/engine/spmd.py
index d334e9f1c95c..96fe41de72b1 100644
--- a/python/cudf_polars/cudf_polars/engine/spmd.py
+++ b/python/cudf_polars/cudf_polars/engine/spmd.py
@@ -10,22 +10,23 @@
 from concurrent.futures import ThreadPoolExecutor
 from typing import TYPE_CHECKING, Any, cast
 
+import pylibcudf as plc
+from cudf_streaming.integrations.partition import (
+    packed_data_from_cudf_packed_columns,
+    unpack_and_concat,
+)
+from pylibcudf.contiguous_split import pack
 from rapidsmpf import bootstrap
 from rapidsmpf.coll import AllGather
 from rapidsmpf.communicator.single import (
     new_communicator as single_communicator,
 )
 from rapidsmpf.communicator.ucxx import barrier
-from rapidsmpf.integrations.cudf.partition import unpack_and_concat
-from rapidsmpf.memory.packed_data import PackedData
 from rapidsmpf.progress_thread import ProgressThread
 from rapidsmpf.rmm_resource_adaptor import RmmResourceAdaptor
 from rapidsmpf.statistics import Statistics
 from rapidsmpf.streaming.core.context import Context
 
-import pylibcudf as plc
-from pylibcudf.contiguous_split import pack
-
 from cudf_polars.containers import DataFrame, DataType
 from cudf_polars.engine.core import (
     ClusterInfo,
@@ -51,11 +52,11 @@
     import uuid
     from collections.abc import Callable
 
+    import polars as pl
+
+    from cudf_streaming.streaming.channel_metadata import ChannelMetadata
     from rapidsmpf.communicator.communicator import Communicator
     from rapidsmpf.config import Options
-    from rapidsmpf.streaming.cudf.channel_metadata import ChannelMetadata
-
-    import polars as pl
 
     from cudf_polars.dsl.ir import IR
     from cudf_polars.engine.core import T
@@ -162,12 +163,11 @@ def allgather_polars_dataframe(
 
     plc_table = plc.Table.from_arrow(local_df, stream=stream)
 
-    packed_data = PackedData.from_cudf_packed_columns(
+    packed_data = packed_data_from_cudf_packed_columns(
         pack(plc_table, stream),
         stream,
         ctx.br(),
     )
-
     # Bulk AllGather: each rank contributes once (sequence_number=0)
     allgather = AllGather(comm, op_id, ctx.br())
     try:
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/allgather.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/allgather.py
index 911e3f270920..8939e50d8121 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/allgather.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/allgather.py
@@ -6,18 +6,18 @@
 
 from typing import TYPE_CHECKING
 
-from rapidsmpf.integrations.cudf.partition import unpack_and_concat
-from rapidsmpf.memory.packed_data import PackedData
-from rapidsmpf.streaming.coll.allgather import AllGather
-
+from cudf_streaming.integrations.partition import (
+    packed_data_from_cudf_packed_columns,
+    unpack_and_concat,
+)
 from pylibcudf.contiguous_split import pack
+from rapidsmpf.streaming.coll.allgather import AllGather
 
 if TYPE_CHECKING:
+    import pylibcudf as plc
+    from cudf_streaming.streaming.table_chunk import TableChunk
     from rapidsmpf.communicator.communicator import Communicator
     from rapidsmpf.streaming.core.context import Context
-    from rapidsmpf.streaming.cudf.table_chunk import TableChunk
-
-    import pylibcudf as plc
     from rmm.pylibrmm.stream import Stream
 
     from cudf_polars.dsl.ir import IRExecutionContext
@@ -72,7 +72,7 @@ def insert(self, sequence_number: int, chunk: TableChunk) -> None:
                 sequence_number,
                 # TODO: Avoid unnecessary copies.
                 # See https://github.com/rapidsai/rapidsmpf/issues/933
-                PackedData.from_cudf_packed_columns(
+                packed_data_from_cudf_packed_columns(
                     pack(
                         chunk.table_view(),
                         chunk.stream,
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/shuffle.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/shuffle.py
index 7da5259afd13..dd78104d2e96 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/shuffle.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/shuffle.py
@@ -6,27 +6,26 @@
 
 from typing import TYPE_CHECKING, Any
 
-from rapidsmpf.communicator.single import new_communicator as single_comm
-from rapidsmpf.config import Options, get_environment_variables
-from rapidsmpf.integrations.cudf.partition import (
+import pylibcudf as plc
+import pylibcudf.partitioning
+from cudf_streaming.integrations.partition import (
     partition_and_pack as py_partition_and_pack,
     split_and_pack as py_split_and_pack,
     unpack_and_concat as py_unpack_and_concat,
 )
+from cudf_streaming.streaming.channel_metadata import (
+    ChannelMetadata,
+    HashScheme,
+    Partitioning,
+)
+from cudf_streaming.streaming.table_chunk import TableChunk
+from rapidsmpf.communicator.single import new_communicator as single_comm
+from rapidsmpf.config import Options, get_environment_variables
 from rapidsmpf.shuffler import PartitionAssignment
 from rapidsmpf.streaming.coll.shuffler import ShufflerAsync
 from rapidsmpf.streaming.core.actor import define_actor
 from rapidsmpf.streaming.core.context import Context
 from rapidsmpf.streaming.core.message import Message
-from rapidsmpf.streaming.cudf.channel_metadata import (
-    ChannelMetadata,
-    HashScheme,
-    Partitioning,
-)
-from rapidsmpf.streaming.cudf.table_chunk import TableChunk
-
-import pylibcudf as plc
-import pylibcudf.partitioning
 
 from cudf_polars.dsl.expr import Col
 from cudf_polars.streaming.actor_graph.dispatch import (
@@ -48,7 +47,6 @@
     from rapidsmpf.communicator.communicator import Communicator
     from rapidsmpf.memory.packed_data import PackedData
     from rapidsmpf.streaming.core.channel import Channel
-
     from rmm.pylibrmm.stream import Stream
 
     from cudf_polars.dsl.ir import IR, IRExecutionContext
@@ -445,7 +443,7 @@ async def shuffle_actor(
 
     This node combines partition_and_pack, shuffler, and unpack_and_concat
     into a single Python node using rapidsmpf.shuffler.Shuffler and utilities
-    from rapidsmpf.integrations.cudf.partition.
+    from cudf_streaming.integrations.partition.
 
     Parameters
     ----------
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/sort.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/sort.py
index 75087141c9b2..f62f2fa05f76 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/sort.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/sort.py
@@ -6,20 +6,19 @@
 
 from typing import TYPE_CHECKING
 
-from rapidsmpf.shuffler import PartitionAssignment
-from rapidsmpf.streaming.core.actor import define_actor
-from rapidsmpf.streaming.core.message import Message
-from rapidsmpf.streaming.cudf.channel_metadata import (
+import polars as pl
+
+import pylibcudf as plc
+from cudf_streaming.streaming.channel_metadata import (
     ChannelMetadata,
     OrderKey,
     OrderScheme,
     Partitioning,
 )
-from rapidsmpf.streaming.cudf.table_chunk import TableChunk
-
-import polars as pl
-
-import pylibcudf as plc
+from cudf_streaming.streaming.table_chunk import TableChunk
+from rapidsmpf.shuffler import PartitionAssignment
+from rapidsmpf.streaming.core.actor import define_actor
+from rapidsmpf.streaming.core.message import Message
 
 from cudf_polars.containers import DataFrame, DataType
 from cudf_polars.dsl.expr import Col, NamedExpr
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/core.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/core.py
index fb7cc4e1439a..1bd440a239f5 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/core.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/core.py
@@ -30,14 +30,14 @@
 if TYPE_CHECKING:
     from collections.abc import MutableMapping
 
+    import polars as pl
+
+    from cudf_streaming.streaming.channel_metadata import ChannelMetadata
+    from cudf_streaming.streaming.table_chunk import TableChunk
     from rapidsmpf.communicator.communicator import Communicator
     from rapidsmpf.streaming.core.channel import Channel
     from rapidsmpf.streaming.core.context import Context
     from rapidsmpf.streaming.core.leaf_actor import DeferredMessages
-    from rapidsmpf.streaming.cudf.channel_metadata import ChannelMetadata
-    from rapidsmpf.streaming.cudf.table_chunk import TableChunk
-
-    import polars as pl
 
     from cudf_polars.dsl.ir import IR, IRExecutionContext
     from cudf_polars.streaming.actor_graph.dispatch import (
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/groupby.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/groupby.py
index c9f65da127c1..df2c8d04f4c7 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/groupby.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/groupby.py
@@ -7,18 +7,17 @@
 from dataclasses import dataclass
 from typing import TYPE_CHECKING, Any
 
+import pylibcudf as plc
+from cudf_streaming.streaming.channel_metadata import (
+    ChannelMetadata,
+    HashScheme,
+)
+from cudf_streaming.streaming.table_chunk import TableChunk
 from rapidsmpf.communicator.single import new_communicator as single_comm
 from rapidsmpf.config import Options, get_environment_variables
 from rapidsmpf.streaming.core.actor import define_actor
 from rapidsmpf.streaming.core.context import Context
 from rapidsmpf.streaming.core.message import Message
-from rapidsmpf.streaming.cudf.channel_metadata import (
-    ChannelMetadata,
-    HashScheme,
-)
-from rapidsmpf.streaming.cudf.table_chunk import TableChunk
-
-import pylibcudf as plc
 
 from cudf_polars.containers import DataType
 from cudf_polars.dsl.expr import Col, NamedExpr
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/io.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/io.py
index 7f4db5e38bb3..f30b55dbdd3d 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/io.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/io.py
@@ -8,15 +8,14 @@
 import math
 from typing import TYPE_CHECKING, Any
 
+import pylibcudf as plc
+from cudf_streaming.streaming.channel_metadata import ChannelMetadata
+from cudf_streaming.streaming.table_chunk import TableChunk
 from rapidsmpf.memory.memory_reservation import opaque_memory_usage
 from rapidsmpf.streaming.core.memory_reserve_or_wait import (
     reserve_memory,
 )
 from rapidsmpf.streaming.core.message import Message
-from rapidsmpf.streaming.cudf.channel_metadata import ChannelMetadata
-from rapidsmpf.streaming.cudf.table_chunk import TableChunk
-
-import pylibcudf as plc
 
 from cudf_polars.dsl.ir import (
     IR,
@@ -481,7 +480,7 @@ def make_rapidsmpf_read_parquet_node(
     The RapidsMPF read parquet node, or None if the predicate cannot be
     converted to a parquet filter (caller should fall back to scan_node).
     """
-    from rapidsmpf.streaming.cudf.parquet import Filter, read_parquet
+    from cudf_streaming.streaming.parquet import Filter, read_parquet
 
     # Build ParquetReaderOptions
     try:
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/join.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/join.py
index 0130146576a6..fa99192aaf12 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/join.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/join.py
@@ -7,25 +7,24 @@
 from dataclasses import dataclass
 from typing import TYPE_CHECKING, Any, Literal
 
-from rapidsmpf.memory.memory_reservation import opaque_memory_usage
-from rapidsmpf.streaming.core.actor import define_actor
-from rapidsmpf.streaming.core.memory_reserve_or_wait import (
-    reserve_memory,
-)
-from rapidsmpf.streaming.core.message import Message
-from rapidsmpf.streaming.cudf.bloom_filter import BloomFilter
-from rapidsmpf.streaming.cudf.channel_metadata import (
+import pylibcudf as plc
+from cudf_streaming.streaming.bloom_filter import BloomFilter
+from cudf_streaming.streaming.channel_metadata import (
     ChannelMetadata,
     HashScheme,
     Partitioning,
 )
-from rapidsmpf.streaming.cudf.table_chunk import (
+from cudf_streaming.streaming.table_chunk import (
     TableChunk,
     make_table_chunks_available_or_wait,
 )
-
-import pylibcudf as plc
 from pylibcudf.hashing import LIBCUDF_DEFAULT_HASH_SEED
+from rapidsmpf.memory.memory_reservation import opaque_memory_usage
+from rapidsmpf.streaming.core.actor import define_actor
+from rapidsmpf.streaming.core.memory_reserve_or_wait import (
+    reserve_memory,
+)
+from rapidsmpf.streaming.core.message import Message
 
 from cudf_polars.containers import DataFrame
 from cudf_polars.dsl.ir import IR, Join
@@ -60,10 +59,10 @@
     from collections.abc import Iterable, MutableMapping
     from types import CoroutineType
 
+    from cudf_streaming.streaming.bloom_filter import BloomFilterChunk
     from rapidsmpf.communicator.communicator import Communicator
     from rapidsmpf.streaming.core.channel import Channel
     from rapidsmpf.streaming.core.context import Context
-    from rapidsmpf.streaming.cudf.bloom_filter import BloomFilterChunk
 
     from cudf_polars.dsl.ir import IR, IRExecutionContext
     from cudf_polars.streaming.actor_graph.dispatch import SubNetGenerator
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/nodes.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/nodes.py
index 9432fd7295d8..777fb02c7a64 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/nodes.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/nodes.py
@@ -7,16 +7,16 @@
 import asyncio
 from typing import TYPE_CHECKING, Any, cast
 
+from cudf_streaming.streaming.channel_metadata import ChannelMetadata
+from cudf_streaming.streaming.table_chunk import (
+    TableChunk,
+    make_table_chunks_available_or_wait,
+)
 from rapidsmpf.memory.buffer import MemoryType
 from rapidsmpf.memory.memory_reservation import opaque_memory_usage
 from rapidsmpf.streaming.core.actor import define_actor
 from rapidsmpf.streaming.core.message import Message
 from rapidsmpf.streaming.core.spillable_messages import SpillableMessages
-from rapidsmpf.streaming.cudf.channel_metadata import ChannelMetadata
-from rapidsmpf.streaming.cudf.table_chunk import (
-    TableChunk,
-    make_table_chunks_available_or_wait,
-)
 
 from cudf_polars.containers import DataFrame
 from cudf_polars.dsl.ir import IR, Empty
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/over.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/over.py
index d1569d561b21..4b0f20db5f7b 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/over.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/over.py
@@ -40,19 +40,18 @@
 from dataclasses import dataclass
 from typing import TYPE_CHECKING, Any, ClassVar, cast
 
+import polars as pl
+
+import pylibcudf as plc
+from cudf_streaming.streaming.channel_metadata import ChannelMetadata
+from cudf_streaming.streaming.table_chunk import (
+    TableChunk,
+    make_table_chunks_available_or_wait,
+)
 from rapidsmpf.memory.memory_reservation import opaque_memory_usage
 from rapidsmpf.shuffler import PartitionAssignment
 from rapidsmpf.streaming.core.actor import define_actor
 from rapidsmpf.streaming.core.message import Message
-from rapidsmpf.streaming.cudf.channel_metadata import ChannelMetadata
-from rapidsmpf.streaming.cudf.table_chunk import (
-    TableChunk,
-    make_table_chunks_available_or_wait,
-)
-
-import polars as pl
-
-import pylibcudf as plc
 
 from cudf_polars.containers import Column, DataFrame, DataType
 from cudf_polars.dsl.expr import GroupedWindow
@@ -92,7 +91,6 @@
     from rapidsmpf.memory.buffer_resource import BufferResource
     from rapidsmpf.streaming.core.channel import Channel
     from rapidsmpf.streaming.core.context import Context
-
     from rmm.pylibrmm.stream import Stream
 
     from cudf_polars.dsl.expr import Col
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/repartition.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/repartition.py
index fcfc897e1728..ccd62b8c1c15 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/repartition.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/repartition.py
@@ -7,14 +7,14 @@
 import math
 from typing import TYPE_CHECKING, Any
 
-from rapidsmpf.memory.memory_reservation import opaque_memory_usage
-from rapidsmpf.streaming.core.actor import define_actor
-from rapidsmpf.streaming.core.message import Message
-from rapidsmpf.streaming.cudf.channel_metadata import ChannelMetadata
-from rapidsmpf.streaming.cudf.table_chunk import (
+from cudf_streaming.streaming.channel_metadata import ChannelMetadata
+from cudf_streaming.streaming.table_chunk import (
     TableChunk,
     make_table_chunks_available_or_wait,
 )
+from rapidsmpf.memory.memory_reservation import opaque_memory_usage
+from rapidsmpf.streaming.core.actor import define_actor
+from rapidsmpf.streaming.core.message import Message
 
 from cudf_polars.containers import DataFrame
 from cudf_polars.streaming.actor_graph.collectives.allgather import AllGatherManager
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/union.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/union.py
index a2664fa3a798..bf14a23f1c15 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/union.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/union.py
@@ -6,9 +6,9 @@
 
 from typing import TYPE_CHECKING, Any
 
+from cudf_streaming.streaming.channel_metadata import ChannelMetadata
+from cudf_streaming.streaming.table_chunk import TableChunk
 from rapidsmpf.streaming.core.message import Message
-from rapidsmpf.streaming.cudf.channel_metadata import ChannelMetadata
-from rapidsmpf.streaming.cudf.table_chunk import TableChunk
 
 from cudf_polars.dsl.ir import Union
 from cudf_polars.streaming.actor_graph.dispatch import (
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/utils.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/utils.py
index ac87b16d5872..68c5e8a94c0d 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/utils.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/utils.py
@@ -16,24 +16,23 @@
 from functools import reduce
 from typing import TYPE_CHECKING, Any, Literal, TypeAlias, cast
 
-from rapidsmpf.memory.memory_reservation import opaque_memory_usage
-from rapidsmpf.memory.packed_data import PackedData
-from rapidsmpf.streaming.coll.allgather import AllGather
-from rapidsmpf.streaming.core.message import Message
-from rapidsmpf.streaming.cudf.channel_metadata import (
+import pylibcudf as plc
+import rmm.mr
+from cudf_streaming.streaming.channel_metadata import (
     ChannelMetadata,
     HashScheme,
     OrderKey,
     OrderScheme,
     Partitioning,
 )
-from rapidsmpf.streaming.cudf.table_chunk import (
+from cudf_streaming.streaming.table_chunk import (
     TableChunk,
     make_table_chunks_available_or_wait,
 )
-
-import pylibcudf as plc
-import rmm.mr
+from rapidsmpf.memory.memory_reservation import opaque_memory_usage
+from rapidsmpf.memory.packed_data import PackedData
+from rapidsmpf.streaming.coll.allgather import AllGather
+from rapidsmpf.streaming.core.message import Message
 
 import cudf_polars.dsl.tracing
 from cudf_polars.containers import DataFrame
@@ -61,7 +60,6 @@
     from rapidsmpf.streaming.core.channel import Channel
     from rapidsmpf.streaming.core.context import Context
     from rapidsmpf.streaming.core.spillable_messages import SpillableMessages
-
     from rmm.pylibrmm.stream import Stream
 
     from cudf_polars.dsl.ir import IR, IRExecutionContext
diff --git a/python/cudf_polars/cudf_polars/utils/config.py b/python/cudf_polars/cudf_polars/utils/config.py
index 4582c0040019..5786a5351cc4 100644
--- a/python/cudf_polars/cudf_polars/utils/config.py
+++ b/python/cudf_polars/cudf_polars/utils/config.py
@@ -36,13 +36,13 @@
     from concurrent.futures import ThreadPoolExecutor
 
     import distributed
-    from rapidsmpf.communicator.communicator import Communicator
-    from rapidsmpf.streaming.core.context import Context
     from ray.actor import ActorHandle
 
     import polars.lazyframe.engine_config
 
     import rmm.mr
+    from rapidsmpf.communicator.communicator import Communicator
+    from rapidsmpf.streaming.core.context import Context
 
     from cudf_polars.engine.ray import RankActor
 
diff --git a/python/cudf_polars/pyproject.toml b/python/cudf_polars/pyproject.toml
index 3f395e4d92ee..557520d9a2fe 100644
--- a/python/cudf_polars/pyproject.toml
+++ b/python/cudf_polars/pyproject.toml
@@ -22,6 +22,7 @@ license-files = ["LICENSE"]
 requires-python = ">=3.11"
 dependencies = [
     "cuda-python>=13.0.1,<14.0",
+    "cudf-streaming==26.8.*,>=0.0.0a0",
     "nvidia-ml-py>=12",
     "packaging",
     "polars>=1.35,<1.42",
@@ -232,7 +233,7 @@ required-imports = ["from __future__ import annotations"]
 
 [tool.ruff.lint.isort.sections]
 polars = ["polars"]
-rapids = ["rmm", "pylibcudf"]
+rapids = ["rmm", "pylibcudf", "rapidsmpf", "cudf_streaming"]
 
 [tool.ruff.format]
 docstring-code-format = true
diff --git a/python/cudf_polars/tests/streaming/test_allgather.py b/python/cudf_polars/tests/streaming/test_allgather.py
index 48d84afdb00f..54cd49b9afff 100644
--- a/python/cudf_polars/tests/streaming/test_allgather.py
+++ b/python/cudf_polars/tests/streaming/test_allgather.py
@@ -8,11 +8,10 @@
 import asyncio
 from concurrent.futures import ThreadPoolExecutor
 
-from rapidsmpf.streaming.cudf.table_chunk import TableChunk
-
 import polars as pl
 
 import pylibcudf as plc
+from cudf_streaming.streaming.table_chunk import TableChunk
 
 from cudf_polars.dsl.ir import IRExecutionContext
 from cudf_polars.streaming.actor_graph.collectives.allgather import AllGatherManager
diff --git a/python/cudf_polars/tests/streaming/test_dask.py b/python/cudf_polars/tests/streaming/test_dask.py
index 3d81ff5cf1c8..c5cabd323aa8 100644
--- a/python/cudf_polars/tests/streaming/test_dask.py
+++ b/python/cudf_polars/tests/streaming/test_dask.py
@@ -9,10 +9,11 @@
 from typing import TYPE_CHECKING
 
 import pytest
-from rapidsmpf.bootstrap import is_running_with_rrun
 
 import polars as pl
 
+from rapidsmpf.bootstrap import is_running_with_rrun
+
 from cudf_polars.engine.options import StreamingOptions
 from cudf_polars.testing.asserts import assert_gpu_result_equal
 from cudf_polars.utils.config import DaskContext
diff --git a/python/cudf_polars/tests/streaming/test_metadata.py b/python/cudf_polars/tests/streaming/test_metadata.py
index 5c555e3593ba..d634c88561a3 100644
--- a/python/cudf_polars/tests/streaming/test_metadata.py
+++ b/python/cudf_polars/tests/streaming/test_metadata.py
@@ -6,18 +6,18 @@
 from __future__ import annotations
 
 import pytest
-from rapidsmpf.streaming.cudf.channel_metadata import (
+
+import polars as pl
+
+import pylibcudf as plc
+from cudf_streaming.streaming.channel_metadata import (
     ChannelMetadata,
     HashScheme,
     OrderKey,
     OrderScheme,
     Partitioning,
 )
-from rapidsmpf.streaming.cudf.table_chunk import TableChunk
-
-import polars as pl
-
-import pylibcudf as plc
+from cudf_streaming.streaming.table_chunk import TableChunk
 
 from cudf_polars import Translator
 from cudf_polars.containers import DataFrame, DataType
diff --git a/python/cudf_polars/tests/streaming/test_ray.py b/python/cudf_polars/tests/streaming/test_ray.py
index a4e543d5c628..3b226eda8147 100644
--- a/python/cudf_polars/tests/streaming/test_ray.py
+++ b/python/cudf_polars/tests/streaming/test_ray.py
@@ -9,10 +9,11 @@
 from unittest.mock import patch
 
 import pytest
-from rapidsmpf.bootstrap import is_running_with_rrun
 
 import polars as pl
 
+from rapidsmpf.bootstrap import is_running_with_rrun
+
 from cudf_polars.engine.hardware_binding import HardwareBindingPolicy
 from cudf_polars.utils.config import RayContext
 
diff --git a/python/cudf_polars/tests/streaming/test_shuffler.py b/python/cudf_polars/tests/streaming/test_shuffler.py
index 7f4fd46d15e1..51ffe28e7f46 100644
--- a/python/cudf_polars/tests/streaming/test_shuffler.py
+++ b/python/cudf_polars/tests/streaming/test_shuffler.py
@@ -6,14 +6,15 @@
 import asyncio
 
 import pytest
-from rapidsmpf.streaming.cudf.channel_metadata import (
+
+import polars as pl
+
+from cudf_streaming.streaming.channel_metadata import (
     ChannelMetadata,
     HashScheme,
     Partitioning,
 )
-from rapidsmpf.streaming.cudf.table_chunk import TableChunk
-
-import polars as pl
+from cudf_streaming.streaming.table_chunk import TableChunk
 
 from cudf_polars.containers import DataFrame, DataType
 from cudf_polars.engine.options import StreamingOptions
diff --git a/python/cudf_polars/tests/streaming/test_spilling.py b/python/cudf_polars/tests/streaming/test_spilling.py
index e368bc3ac1f3..c1d7e5b1cdfc 100644
--- a/python/cudf_polars/tests/streaming/test_spilling.py
+++ b/python/cudf_polars/tests/streaming/test_spilling.py
@@ -9,15 +9,15 @@
 from typing import TYPE_CHECKING
 
 import pytest
-from rapidsmpf.memory.buffer import MemoryType
-from rapidsmpf.memory.pinned_memory_resource import is_pinned_memory_resources_supported
-from rapidsmpf.streaming.core.message import Message
-from rapidsmpf.streaming.core.spillable_messages import SpillableMessages
-from rapidsmpf.streaming.cudf.table_chunk import TableChunk
 
 import polars as pl
 
 import pylibcudf as plc
+from cudf_streaming.streaming.table_chunk import TableChunk
+from rapidsmpf.memory.buffer import MemoryType
+from rapidsmpf.memory.pinned_memory_resource import is_pinned_memory_resources_supported
+from rapidsmpf.streaming.core.message import Message
+from rapidsmpf.streaming.core.spillable_messages import SpillableMessages
 
 from cudf_polars.engine.options import StreamingOptions
 from cudf_polars.streaming.actor_graph.utils import (
diff --git a/python/cudf_polars/tests/streaming/test_spmd.py b/python/cudf_polars/tests/streaming/test_spmd.py
index 17c2f86d961a..ed4a5b069ba9 100644
--- a/python/cudf_polars/tests/streaming/test_spmd.py
+++ b/python/cudf_polars/tests/streaming/test_spmd.py
@@ -9,12 +9,12 @@
 from unittest.mock import patch
 
 import pytest
-from rapidsmpf.bootstrap import is_running_with_rrun
-from rapidsmpf.rmm_resource_adaptor import RmmResourceAdaptor
 
 import polars as pl
 
 import rmm.mr
+from rapidsmpf.bootstrap import is_running_with_rrun
+from rapidsmpf.rmm_resource_adaptor import RmmResourceAdaptor
 
 from cudf_polars.engine.core import _find_memory_error
 from cudf_polars.engine.hardware_binding import HardwareBindingPolicy
diff --git a/python/cudf_polars/tests/streaming/test_statistics.py b/python/cudf_polars/tests/streaming/test_statistics.py
index 61d5659dc702..f51605d6bfd7 100644
--- a/python/cudf_polars/tests/streaming/test_statistics.py
+++ b/python/cudf_polars/tests/streaming/test_statistics.py
@@ -7,6 +7,7 @@
 from typing import TYPE_CHECKING
 
 import pytest
+
 from rapidsmpf.statistics import Statistics
 
 from cudf_polars.engine.options import StreamingOptions
diff --git a/python/cudf_streaming/cudf_streaming/integrations/__init__.py b/python/cudf_streaming/cudf_streaming/integrations/__init__.py
index 743c0b3ec454..37f3e8a1575c 100644
--- a/python/cudf_streaming/cudf_streaming/integrations/__init__.py
+++ b/python/cudf_streaming/cudf_streaming/integrations/__init__.py
@@ -2,6 +2,9 @@
 # SPDX-License-Identifier: Apache-2.0
 """Collection of cuDF specific functions."""
 
-from cudf_streaming.integrations.partition import unpack_and_concat
+from cudf_streaming.integrations.partition import (
+    packed_data_from_cudf_packed_columns,
+    unpack_and_concat,
+)
 
-__all__ = ["unpack_and_concat"]
+__all__ = ["packed_data_from_cudf_packed_columns", "unpack_and_concat"]
diff --git a/python/cudf_streaming/cudf_streaming/integrations/partition.pxd b/python/cudf_streaming/cudf_streaming/integrations/partition.pxd
index 774ccb338c10..9d674b1bd4c0 100644
--- a/python/cudf_streaming/cudf_streaming/integrations/partition.pxd
+++ b/python/cudf_streaming/cudf_streaming/integrations/partition.pxd
@@ -1,6 +1,7 @@
 # SPDX-FileCopyrightText: Copyright (c) 2025-2026, NVIDIA CORPORATION & AFFILIATES.
 # SPDX-License-Identifier: Apache-2.0
 
+from pylibcudf.contiguous_split cimport PackedColumns
 from pylibcudf.table cimport Table
 from rmm.pylibrmm.stream cimport Stream
 
@@ -27,3 +28,8 @@ cpdef object unspill_partitions(
     BufferResource br,
     object allow_overbooking,
 )
+cpdef object packed_data_from_cudf_packed_columns(
+    PackedColumns packed_columns,
+    Stream stream,
+    BufferResource br,
+)
diff --git a/python/cudf_streaming/cudf_streaming/integrations/partition.pyi b/python/cudf_streaming/cudf_streaming/integrations/partition.pyi
index 1133c65ab1c9..efb30be09176 100644
--- a/python/cudf_streaming/cudf_streaming/integrations/partition.pyi
+++ b/python/cudf_streaming/cudf_streaming/integrations/partition.pyi
@@ -4,6 +4,7 @@ from __future__ import annotations
 
 from collections.abc import Iterable
 
+from pylibcudf.contiguous_split import PackedColumns
 from pylibcudf.table import Table
 
 from rapidsmpf.memory.buffer_resource import BufferResource
@@ -39,3 +40,8 @@ def unspill_partitions(
     br: BufferResource,
     allow_overbooking: bool,
 ) -> list[PackedData]: ...
+def packed_data_from_cudf_packed_columns(
+    packed_columns: PackedColumns,
+    stream: Stream,
+    br: BufferResource,
+) -> PackedData: ...
diff --git a/python/cudf_streaming/cudf_streaming/integrations/partition.pyx b/python/cudf_streaming/cudf_streaming/integrations/partition.pyx
index cc607d841934..f9fdc5b62203 100644
--- a/python/cudf_streaming/cudf_streaming/integrations/partition.pyx
+++ b/python/cudf_streaming/cudf_streaming/integrations/partition.pyx
@@ -4,16 +4,18 @@
 
 from cython.operator cimport dereference as deref
 from cython.operator cimport postincrement
-from libc.stdint cimport uint32_t
+from libc.stdint cimport uint8_t, uint32_t
 from libcpp.memory cimport make_unique, unique_ptr
 from libcpp.unordered_map cimport unordered_map
 from libcpp.utility cimport move
 from libcpp.vector cimport vector
+from pylibcudf.contiguous_split cimport PackedColumns
 from pylibcudf.libcudf.table.table cimport table as cpp_table
 from pylibcudf.libcudf.table.table_view cimport table_view
 from pylibcudf.libcudf.types cimport size_type
 from pylibcudf.table cimport Table
 from rmm.librmm.cuda_stream_view cimport cuda_stream_view
+from rmm.librmm.device_buffer cimport device_buffer
 from rmm.pylibrmm.stream cimport Stream
 
 from rapidsmpf._detail.exception_handling cimport ex_handler
@@ -316,6 +318,32 @@ cdef extern from "<cudf_streaming/integrations/partition.hpp>" nogil:
         ) except +ex_handler
 
 
+cdef extern from *:
+    """
+    #include <rapidsmpf/error.hpp>
+    #include <rapidsmpf/memory/packed_data.hpp>
+    #include <rapidsmpf/memory/buffer_resource.hpp>
+    #include <rmm/device_buffer.hpp>
+
+    std::unique_ptr<rapidsmpf::PackedData> cpp_packed_data_from_buffers(
+        std::unique_ptr<std::vector<std::uint8_t>> metadata,
+        std::unique_ptr<rmm::device_buffer> gpu_data,
+        rmm::cuda_stream_view stream,
+        rapidsmpf::BufferResource* br
+    ) {
+        return std::make_unique<rapidsmpf::PackedData>(
+            std::move(metadata), br->move(std::move(gpu_data), stream)
+        );
+    }
+    """
+    unique_ptr[cpp_PackedData] cpp_packed_data_from_buffers(
+        unique_ptr[vector[uint8_t]] metadata,
+        unique_ptr[device_buffer] gpu_data,
+        cuda_stream_view stream,
+        cpp_BufferResource* br,
+    ) except +ex_handler nogil
+
+
 cpdef object unspill_partitions(
     object partitions,
     BufferResource br,
@@ -369,3 +397,59 @@ cpdef object unspill_partitions(
             ab,
         )
     return packed_data_vector_to_list(move(_ret), br)
+
+
+cpdef object packed_data_from_cudf_packed_columns(
+    PackedColumns packed_columns,
+    Stream stream,
+    BufferResource br,
+):
+    """
+    Construct a PackedData from a pylibcudf PackedColumns.
+
+    Takes ownership of the metadata and GPU data from the PackedColumns
+    object, leaving it empty.
+
+    Parameters
+    ----------
+    packed_columns
+        Packed columns from ``pylibcudf.contiguous_split.pack()``.
+        Must not already be empty (already released).
+    stream
+        The CUDA stream on which the preceding ``pack()`` call was performed.
+        Must be the same stream to ensure correct memory ordering.
+    br
+        Buffer resource for memory management.
+
+    Returns
+    -------
+    A new PackedData instance owning the packed column data.
+
+    Raises
+    ------
+    ValueError
+        If the PackedColumns object is empty (already released).
+
+    See Also
+    --------
+    pylibcudf.contiguous_split.pack
+    cudf_streaming.integrations.partition.unpack_and_concat
+    """
+    if packed_columns is None or stream is None or br is None:
+        raise TypeError("Arguments must not be None")
+    cdef cuda_stream_view _stream = stream.view()
+    cdef cpp_BufferResource* _br = br.ptr()
+    cdef PackedData ret = PackedData.__new__(PackedData)
+    with nogil:
+        if not (packed_columns.c_obj != NULL and
+                deref(packed_columns.c_obj).metadata and
+                deref(packed_columns.c_obj).gpu_data):
+            raise ValueError("Cannot release empty PackedColumns")
+        ret.c_obj = cpp_packed_data_from_buffers(
+            move(deref(packed_columns.c_obj).metadata),
+            move(deref(packed_columns.c_obj).gpu_data),
+            _stream,
+            _br,
+        )
+    ret._br = br
+    return ret
diff --git a/python/cudf_streaming/cudf_streaming/tests/test_table_chunk.py b/python/cudf_streaming/cudf_streaming/tests/test_table_chunk.py
index f48a83e84131..6b55abd01f3c 100644
--- a/python/cudf_streaming/cudf_streaming/tests/test_table_chunk.py
+++ b/python/cudf_streaming/cudf_streaming/tests/test_table_chunk.py
@@ -9,6 +9,9 @@
 import pylibcudf as plc
 import pytest
 
+from cudf_streaming.integrations.partition import (
+    packed_data_from_cudf_packed_columns,
+)
 from cudf_streaming.streaming.table_chunk import (
     TableChunk,
     make_table_chunks_available_or_wait,
@@ -406,7 +409,7 @@ def test_shape_accessor(
     expected_shape = (expect.num_rows(), expect.num_columns())
 
     if from_pack:
-        pd = PackedData.from_cudf_packed_columns(
+        pd = packed_data_from_cudf_packed_columns(
             plc.contiguous_split.pack(expect, stream), stream, context.br()
         )
         device_chunk = TableChunk.from_packed_data(pd, br=context.br())
@@ -444,7 +447,7 @@ def test_into_packed_data(
 ) -> None:
     expect = random_table(1024)
     if from_pack:
-        pd = PackedData.from_cudf_packed_columns(
+        pd = packed_data_from_cudf_packed_columns(
             plc.contiguous_split.pack(expect, stream), stream, context.br()
         )
         chunk = TableChunk.from_packed_data(pd, br=context.br())