diff --git a/.github/workflows/hamilton-core-main.yml b/.github/workflows/hamilton-core-main.yml
new file mode 100644
index 000000000..f26ec478a
--- /dev/null
+++ b/.github/workflows/hamilton-core-main.yml
@@ -0,0 +1,48 @@
+name: Unit tests (hamilton-core)
+
+on:
+  workflow_dispatch:
+
+  pull_request:
+    branches:
+        - main
+    paths:
+        - '.github/**'
+        - 'hamilton/**'
+        - 'tests/**'
+        - 'pyproject.toml'
+
+jobs:
+  test:
+    name: "Unit Tests (hamilton-core)"
+    runs-on: ubuntu-latest
+    env:
+      UV_PRERELEASE: "allow"
+      HAMILTON_TELEMETRY_ENABLED: false
+
+    steps:
+        - name: Install Graphviz on Linux
+          if: runner.os == 'Linux'
+          run: sudo apt-get update && sudo apt-get install --yes --no-install-recommends graphviz
+
+        - name: Checkout repository
+          uses: actions/checkout@v4
+
+        - name: Install uv and set the python version
+          uses: astral-sh/setup-uv@v6
+          with:
+            python-version: "3.12"  # most popular Python version
+            enable-cache: true
+            cache-dependency-glob: "uv.lock"
+            activate-environment: true
+
+        - name: Install dependencies
+          run: |
+            uv venv
+            . .venv/bin/activate
+            uv pip install ./hamilton-core[core-tests]
+
+        # NOTE `test_caching.py` is the older caching mechanism
+        - name: Test hamilton main package
+          run: |
+            uv run pytest tests/ --ignore tests/integrations --ignore tests/plugins --ignore tests/test_caching.py
diff --git a/hamilton-core/.gitignore b/hamilton-core/.gitignore
new file mode 100644
index 000000000..750d8503c
--- /dev/null
+++ b/hamilton-core/.gitignore
@@ -0,0 +1 @@
+hamilton/_hamilton
diff --git a/hamilton-core/README.md b/hamilton-core/README.md
new file mode 100644
index 000000000..52accb309
--- /dev/null
+++ b/hamilton-core/README.md
@@ -0,0 +1,41 @@
+# Read carefully
+
+> Use at your own risk
+
+This directory contains code for the package `sf-hamilton-core`. It is a drop-in replacement of `sf-hamilton`, with two changes:
+- disable plugin autoloading
+- make `pandas` and `numpy` optional dependencies; and remove `networkx` dependency (currently unused).
+
+This makes the Hamilton package a much lighter install and solves long library loading time.
+
+## As a user
+If you want to try `sf-hamilton-core`, you need to:
+1. Remove your current Hamilton installation: `pip uninstall sf-hamilton`
+2. Install Hamilton core `pip install sf-hamilton-core`
+3. Check installation `pip list` should only include `sf-hamilton-core`.
+
+This will install a different Python package with the name `hamilton` with the smaller dependencies and plugin autoloading disabled.
+
+It should be a drop-in replacement and your existing Hamilton code should just work. Though, if you're relying on plugins (e.g., parquet materializers, dataframe result builders), you will need to manually load them.
+
+
+## How does it work
+
+
+## Why is another package `sf-hamilton` necessary
+This exists to prevent backwards incompatible changes for people who `pip install sf-hamilton` and use it in production. It is a temporary solution until a major release `sf-hamilton==2.0.0` could allow breaking changes and a more robust solution.
+
+### Disable plugin autoloading
+Hamilton has generous number of plugins (`pandas`, `polars`, `mlflow`, `spark`). To give a good user experience, Hamilton autoloads plugins based on the available Python libraries in the current Python environment. For example, `to.mlflow()` becomes available if `mlflow` is installed. Autoloaded features notably include materializers like `from_.parquet` and `to.parquet` and data validators (pydantic, pandera, etc.)
+
+The issue with this approach is that Python environment with a lot of dependencies, common in data science, can be very slow to start because of all the imports. Currently, Hamilton allows to disable autoloading via a user config or Python code. This require manual setups and is not the best default for some users.
+
+### `pandas` and `numpy` dependencies
+Hamilton was initially created for workflows that used `pandas` and `numpy` heavily. For this reason, `numpy` and `pandas` are imported at the top-level of module `hamilton.base`. Because of the package structure, as a Hamilton user, you're importing `pandas` and `numpy` every time you import `hamilton`.
+
+A reasonable change would be to move `numpy` and `pandas` to a "lazy" location. Then, dependencies would only be imported when features requiring them are used and they could be removed from `pyproject.toml`. Unfortunately, plugin autoloading defaults make this solution a significant breaking change and insatisfactory.
+
+Since plugins are loaded based on the Python package available, removing `pandas` and `numpy` would allow disable the loading of these plugins. This would break popular CSV and parquet materializers.
+
+### `networkx` dependency
+The `sf-hamilton[visualization]` extra currently includes `networkx` as a dependency, though it is never actually used. There's a single function requiring it and it could be implemented in pure Python. This has been made even easier with the addition of `graphlib` in the standard library in Python 3.9.
diff --git a/hamilton-core/hamilton/__init__.py b/hamilton-core/hamilton/__init__.py
new file mode 100644
index 000000000..469c2b01d
--- /dev/null
+++ b/hamilton-core/hamilton/__init__.py
@@ -0,0 +1,70 @@
+import importlib.util
+import pathlib
+import sys
+from types import ModuleType
+from typing import Any
+
+
+def _load_hamilton_module() -> ModuleType:
+    """Patch this relative import in the Hamilton core repository
+
+    ```python
+    # hamilton/__init__.py
+    try:
+        from .version import VERSION as __version__  # noqa: F401
+    except ImportError:
+        from version import VERSION as __version__  # noqa: F401
+    ```
+    """
+
+    origin_path = pathlib.Path(__file__).parent / "_hamilton" / "__init__.py"
+    origin_spec = importlib.util.spec_from_file_location("hamilton", origin_path)
+    origin_module = importlib.util.module_from_spec(origin_spec)
+
+    # The following lines are only required if we don't modify `hamilton/__init__.py`
+    # source_segment = "from version import VERSION as __version__"
+    # # the namespace `hamilton._hamilton` is only temporarily available; it will be removed
+    # # by the end of this initialization
+    # patched_segment = "from hamilton._hamilton.version import VERSION as __version__"
+
+    # source_code = pathlib.Path(origin_path).read_text()
+    # patched_code = source_code.replace(source_segment, patched_segment)
+
+    # exec(patched_code, origin_module.__dict__)
+    # sys.modules["hamilton"] = origin_module
+
+    origin_spec.loader.exec_module(origin_module)
+    return origin_module
+
+
+def _load_hamilton_registry_module():
+    module_path = pathlib.Path(__file__).parent / "_hamilton" / "registry.py"
+    module_spec = importlib.util.spec_from_file_location("hamilton.registry", module_path)
+    module = importlib.util.module_from_spec(module_spec)
+    module_spec.loader.exec_module(module)
+    return module
+
+
+def _create_proxy_module() -> ModuleType:
+    proxy_module = ModuleType(__name__)
+    sys.modules[__name__] = proxy_module
+    return proxy_module
+
+
+_registry_module = _load_hamilton_registry_module()
+# disable plugin autoloading
+_registry_module.disable_autoload()
+
+_origin_module = _load_hamilton_module()
+_proxy_module = _create_proxy_module()
+
+
+def __getattr__(name: str) -> Any:
+    try:
+        return getattr(_origin_module, name)
+    except AttributeError:
+        raise
+
+
+# `getattr()` must be available to build the package
+_proxy_module.__getattr__ = __getattr__
diff --git a/hamilton-core/setup.py b/hamilton-core/setup.py
new file mode 100644
index 000000000..264eaea83
--- /dev/null
+++ b/hamilton-core/setup.py
@@ -0,0 +1,95 @@
+import os
+import pathlib
+import re
+import shutil
+import sys
+
+import tomllib
+from setuptools import setup
+
+os.chdir(os.path.abspath(os.path.dirname(__file__)))
+
+
+def copy_hamilton_library():
+    setup_dir = pathlib.Path(__file__).resolve().parent
+    source_dir = (setup_dir.parent / "hamilton").resolve()
+    dest_dir = (setup_dir / "hamilton" / "_hamilton").resolve()
+
+    # Safety checks
+    if not source_dir.is_dir():
+        print(f"Error: Source directory does not exist: {source_dir}")
+        sys.exit(1)
+
+    if not str(dest_dir).startswith(str(setup_dir)):
+        print(f"Error: Destination directory {dest_dir} is outside the setup directory {setup_dir}")
+        sys.exit(1)
+
+    # Remove destination if it exists to avoid errors or stale files
+    if dest_dir.exists():
+        print("delete: ", dest_dir)
+        shutil.rmtree(dest_dir)
+
+    # Copy entire directory tree from source to destination
+    print(f"copy from: {source_dir}; to {dest_dir}")
+    shutil.copytree(source_dir, dest_dir)
+
+
+def get_version():
+    version_path = pathlib.Path(__file__).parent / "hamilton" / "_hamilton" / "version.py"
+    content = version_path.read_text()
+    match = re.search(r"^VERSION\s*=\s*\(([^)]+)\)", content, re.MULTILINE)
+    if match:
+        version_tuple_str = match.group(1)  # "1, 88, 0"
+        # Parse tuple string into list of integers
+        version_parts = [part.strip() for part in version_tuple_str.split(",")]
+        version_str = ".".join(version_parts)
+        return version_str
+
+
+copy_hamilton_library()
+
+pyproject_path = pathlib.Path(__file__).parents[1] / "pyproject.toml"
+pyproject = tomllib.loads(pyproject_path.read_text())
+project = pyproject["project"]
+
+readme_file = project.get("readme", None)
+console_scripts = [
+    f"{name}={target}"
+    for name, target in project.get("entry-points", {}).get("console_scripts", {}).items()
+]
+install_requires = list(set(project.get("dependencies", [])).difference(set(["pandas", "numpy"])))
+extras_require = {
+    **project.get("optional-dependencies", {}),
+    **{"visualization": ["graphviz"]},  # drop networkx
+    **{
+        "core-tests": [  # dependencies required to run unit tests; used in CI
+            "pytest",
+            "pytest-asyncio",
+            "pandas",
+            "typer",
+            "networkx",
+            "graphviz",
+        ]
+    },
+}
+
+
+setup(
+    name="sf-hamilton-core",
+    version=get_version(),
+    description=project.get("description", ""),
+    long_description=pathlib.Path(readme_file).read_text() if readme_file else "",
+    long_description_content_type="text/markdown" if readme_file else None,
+    python_requires=project.get("requires-python", None),
+    license=project.get("license", {}).get("text", None),
+    keywords=project.get("keywords", []),
+    author=", ".join(a["name"] for a in project.get("authors", [])),
+    author_email=", ".join(a["email"] for a in project.get("authors", [])),
+    classifiers=project.get("classifiers", []),
+    install_requires=install_requires,
+    extras_require=extras_require,
+    entry_points={"console_scripts": console_scripts},
+    project_urls=project.get("urls", {}),
+    packages=["hamilton"],
+    package_data={"hamilton": ["*.json", "*.md", "*.txt"]},
+)
diff --git a/hamilton/__init__.py b/hamilton/__init__.py
index a302407d2..4d3fd39f9 100644
--- a/hamilton/__init__.py
+++ b/hamilton/__init__.py
@@ -1,7 +1,7 @@
 try:
     from .version import VERSION as __version__  # noqa: F401
 except ImportError:
-    from version import VERSION as __version__  # noqa: F401
+    from hamilton.version import VERSION as __version__  # noqa: F401
 
 # this supposedly is required for namespace packages to work.
 __path__ = __import__("pkgutil").extend_path(__path__, __name__)
diff --git a/hamilton/base.py b/hamilton/base.py
index 02241aaca..77fb46a9d 100644
--- a/hamilton/base.py
+++ b/hamilton/base.py
@@ -20,21 +20,22 @@
 It cannot import hamilton.graph, or hamilton.driver.
 """
 
+from __future__ import annotations
+
 import abc
 import collections
 import logging
-from typing import Any, Dict, List, Optional, Tuple, Type, Union
-
-import numpy as np
-import pandas as pd
-from pandas.core.indexes import extension as pd_extension
+from typing import TYPE_CHECKING, Any, Dict, List, Optional, Tuple, Type, Union
 
+from hamilton import htypes
 from hamilton.lifecycle import api as lifecycle_api
 
-try:
-    from . import htypes, node
-except ImportError:
-    import node
+if TYPE_CHECKING:
+    import numpy as np
+    import pandas as pd
+
+    import hamilton.node as node
+
 
 logger = logging.getLogger(__name__)
 
@@ -120,6 +121,8 @@ def pandas_index_types(
         :param outputs: the dict we're trying to create a result from.
         :return: dict of all index types, dict of time series/categorical index types, dict if there is no index
         """
+        import pandas as pd
+
         all_index_types = collections.defaultdict(list)
         time_indexes = collections.defaultdict(list)
         no_indexes = collections.defaultdict(list)
@@ -131,6 +134,8 @@ def index_key_name(pd_object: Union[pd.DataFrame, pd.Series]) -> str:
 
         def get_parent_time_index_type():
             """Helper to pull the right time index parent class."""
+            from pandas.core.indexes import extension as pd_extension
+
             if hasattr(pd_extension, "NDArrayBackedExtensionIndex"):
                 index_type = pd_extension.NDArrayBackedExtensionIndex
             else:
@@ -220,6 +225,8 @@ def build_result(**outputs: Dict[str, Any]) -> pd.DataFrame:
 
         :param outputs: the outputs to build a dataframe from.
         """
+        import pandas as pd
+
         # TODO check inputs are pd.Series, arrays, or scalars -- else error
         output_index_type_tuple = PandasDataFrameResult.pandas_index_types(outputs)
         # this next line just log warnings
@@ -255,6 +262,7 @@ def build_dataframe_with_dataframes(outputs: Dict[str, Any]) -> pd.DataFrame:
         :param outputs: The outputs to build the dataframe from.
         :return: A dataframe with the outputs.
         """
+        import pandas as pd
 
         def get_output_name(output_name: str, column_name: str) -> str:
             """Add function prefix to columns.
@@ -300,6 +308,8 @@ def input_types(self) -> List[Type[Type]]:
         return [Any]
 
     def output_type(self) -> Type:
+        import pandas as pd
+
         return pd.DataFrame
 
 
@@ -365,6 +375,8 @@ def build_result(**outputs: Dict[str, Any]) -> np.matrix:
         :param outputs: function_name -> np.array.
         :return: numpy matrix
         """
+        import numpy as np
+
         # TODO check inputs are all numpy arrays/array like things -- else error
         num_rows = -1
         columns_with_lengths = collections.OrderedDict()
@@ -402,6 +414,8 @@ def input_types(self) -> List[Type[Type]]:
         return [Any]  # Typing
 
     def output_type(self) -> Type:
+        import pandas as pd
+
         return pd.DataFrame