diff --git a/ci/test_wheel_cuvs.sh b/ci/test_wheel_cuvs.sh
index eae78b4a89..c0758da20b 100755
--- a/ci/test_wheel_cuvs.sh
+++ b/ci/test_wheel_cuvs.sh
@@ -13,8 +13,18 @@ RAPIDS_PY_CUDA_SUFFIX="$(rapids-wheel-ctk-name-gen "${RAPIDS_CUDA_VERSION}")"
 LIBCUVS_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="libcuvs_${RAPIDS_PY_CUDA_SUFFIX}" rapids-download-wheels-from-github cpp)
 CUVS_WHEELHOUSE=$(rapids-download-from-github "$(rapids-package-name "wheel_python" cuvs --stable --cuda "$RAPIDS_CUDA_VERSION")")
 
-# echo to expand wildcard before adding `[extra]` requires for pip
+# generate constraints (possibly pinning to oldest support versions of dependencies)
+rapids-generate-pip-constraints test_python "${PIP_CONSTRAINT}"
+
+# notes:
+#
+#   * echo to expand wildcard before adding `[test]` requires for pip
+#   * just providing --constraint="${PIP_CONSTRAINT}" to be explicit, and because
+#     that environment variable is ignored if any other --constraint are passed via the CLI
+#
 rapids-pip-retry install \
+    --prefer-binary \
+    --constraint "${PIP_CONSTRAINT}" \
     "${LIBCUVS_WHEELHOUSE}"/libcuvs*.whl \
     "$(echo "${CUVS_WHEELHOUSE}"/cuvs*.whl)[test]"
 
diff --git a/conda/recipes/libcuvs/recipe.yaml b/conda/recipes/libcuvs/recipe.yaml
index abd3031a94..c598bb59ce 100644
--- a/conda/recipes/libcuvs/recipe.yaml
+++ b/conda/recipes/libcuvs/recipe.yaml
@@ -134,9 +134,6 @@ outputs:
         - libcurand
         - libcusolver
         - libcusparse
-        - if: cuda_major == "13"
-          then:
-            - libnvjitlink
       ignore_run_exports:
         by_name:
           - cuda-cudart
@@ -150,9 +147,6 @@ outputs:
           - librmm
           - mkl
           - nccl
-          - if: cuda_major == "13"
-            then:
-              - libnvjitlink
     about:
       homepage: ${{ load_from_file("python/libcuvs/pyproject.toml").project.urls.Homepage }}
       license: ${{ load_from_file("python/libcuvs/pyproject.toml").project.license }}
@@ -203,9 +197,6 @@ outputs:
         - libcurand
         - libcusolver
         - libcusparse
-        - if: cuda_major == "13"
-          then:
-            - libnvjitlink
       ignore_run_exports:
         by_name:
           - cuda-cudart
@@ -219,9 +210,6 @@ outputs:
           - librmm
           - mkl
           - nccl
-          - if: cuda_major == "13"
-            then:
-              - libnvjitlink
     about:
       homepage: ${{ load_from_file("python/libcuvs/pyproject.toml").project.urls.Homepage }}
       license: ${{ load_from_file("python/libcuvs/pyproject.toml").project.license }}
@@ -270,9 +258,6 @@ outputs:
         - libcurand
         - libcusolver
         - libcusparse
-        - if: cuda_major == "13"
-          then:
-            - libnvjitlink
       ignore_run_exports:
         by_name:
           - cuda-cudart
@@ -283,9 +268,6 @@ outputs:
           - libcurand
           - libcusolver
           - libcusparse
-          - if: cuda_major == "13"
-            then:
-              - libnvjitlink
           - librmm
           - mkl
           - nccl
@@ -436,9 +418,6 @@ outputs:
         - libcurand
         - libcusolver
         - libcusparse
-        - if: cuda_major == "13"
-          then:
-            - libnvjitlink
       ignore_run_exports:
         by_name:
           - cuda-cudart
@@ -449,9 +428,6 @@ outputs:
           - libcurand
           - libcusolver
           - libcusparse
-          - if: cuda_major == "13"
-            then:
-              - libnvjitlink
           - librmm
           - mkl
           - nccl
diff --git a/dependencies.yaml b/dependencies.yaml
index 2c093251e0..6cc68b8dda 100644
--- a/dependencies.yaml
+++ b/dependencies.yaml
@@ -286,6 +286,8 @@ dependencies:
           - clang==20.1.4
           - clang-tools==20.1.4
           - libclang==20.1.4
+  # 'cuda_version' intentionally does not contain fallback entries... we want
+  # a loud error if an unsupported 'cuda' value is passed
   cuda_version:
     specific:
       - output_types: conda
@@ -314,6 +316,51 @@ dependencies:
               cuda: "13.1"
             packages:
               - cuda-version=13.1
+      - output_types: requirements
+        matrices:
+          # if use_cuda_wheels=false is provided, do not add dependencies on any CUDA wheels
+          # (e.g. for DLFW and pip devcontainers)
+          - matrix:
+              use_cuda_wheels: "false"
+            packages:
+          - matrix:
+              arch: aarch64
+              cuda: "12.2"
+              use_cuda_wheels: "true"
+            packages:
+              # some components (like nvidia-cublas-cu12 and nvidia-cuda-nvcc-cu12) didn't have
+              # aarch64 wheels until CTK 12.3, so allow a slightly looser bound here
+              - cuda-toolkit>=12.2,<12.4
+          - matrix:
+              cuda: "12.2"
+              use_cuda_wheels: "true"
+            packages:
+              - cuda-toolkit==12.2.*
+          - matrix:
+              cuda: "12.5"
+              use_cuda_wheels: "true"
+            packages:
+              - cuda-toolkit==12.5.*
+          - matrix:
+              cuda: "12.8"
+              use_cuda_wheels: "true"
+            packages:
+              - cuda-toolkit==12.8.*
+          - matrix:
+              cuda: "12.9"
+              use_cuda_wheels: "true"
+            packages:
+              - cuda-toolkit==12.9.*
+          - matrix:
+              cuda: "13.0"
+              use_cuda_wheels: "true"
+            packages:
+              - cuda-toolkit==13.0.*
+          - matrix:
+              cuda: "13.1"
+              use_cuda_wheels: "true"
+            packages:
+              - cuda-toolkit==13.1.*
   cuda:
     common:
       - output_types: [conda]
@@ -337,26 +384,40 @@ dependencies:
             packages:
   cuda_wheels:
     specific:
+      # cuVS needs 'nvJitLink>={whatever-cuvs-was-built-against}' at runtime, and mixing
+      # old-CTK with new-nvJitLink is supported, so maintain 1 entry here per CUDA major.minor
+      # cuVS is built against.
+      #
+      # nvJitLink has historically roughly followed the CTK's minor versioning scheme,
+      # but check at https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html
       - output_types: [requirements, pyproject]
         matrices:
+          # skip this whole group if explicitly opting out of CUDA wheels
           - matrix:
-              cuda: "12.*"
-              use_cuda_wheels: "true"
+              use_cuda_wheels: "false"
             packages:
-              - cuda-toolkit[cublas,curand,cusolver,cusparse,nvjitlink]==12.*
+          # We always want nvJitLink >= whatever minor version was available in the build environment
+          # nvJitLink tends to follow the CTK's minor versions, but check at https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html
+          #
+          # ref: https://github.com/rapidsai/cudf/pull/20873
           - matrix:
-              cuda: "13.*"
+              cuda: "12.9"
               use_cuda_wheels: "true"
             packages:
-              - cuda-toolkit[cublas,curand,cusolver,cusparse,nvjitlink]==13.*
+              - cuda-toolkit[cublas,curand,cusolver,cusparse]==12.*
+              - nvidia-nvjitlink-cu12>=12.9,<13
           - matrix:
-              use_cuda_wheels: "false"
+              cuda: "13.*"
+              use_cuda_wheels: "true"
             packages:
-          # if no matching matrix selectors passed, list a range
+              - &ctk_cu13 cuda-toolkit[cublas,curand,cusolver,cusparse]==13.*
+              - &nvjitlink_cu13 nvidia-nvjitlink>=13.0,<14
+          # if no matching matrix selectors passed, list the CUDA 13 requirement
           # (just as a source of documentation, as this populates pyproject.toml in source control)
           - matrix:
             packages:
-              - cuda-toolkit[cublas,curand,cusolver,cusparse,nvjitlink]>=12,<14
+              - *ctk_cu13
+              - *nvjitlink_cu13
   depends_on_cupy:
     common:
       - output_types: conda
diff --git a/python/libcuvs/pyproject.toml b/python/libcuvs/pyproject.toml
index f43bc35dbf..052834cab0 100644
--- a/python/libcuvs/pyproject.toml
+++ b/python/libcuvs/pyproject.toml
@@ -19,9 +19,10 @@ authors = [
 license = "Apache-2.0"
 requires-python = ">=3.11"
 dependencies = [
-    "cuda-toolkit[cublas,curand,cusolver,cusparse,nvjitlink]>=12,<14",
+    "cuda-toolkit[cublas,curand,cusolver,cusparse]==13.*",
     "libraft==26.4.*,>=0.0.0a0",
     "librmm==26.4.*,>=0.0.0a0",
+    "nvidia-nvjitlink>=13.0,<14",
 ] # This list was generated by `rapids-dependency-file-generator`. To make changes, edit ../../dependencies.yaml and run `rapids-dependency-file-generator`.
 classifiers = [
     "Intended Audience :: Developers",