diff --git a/setup.cfg b/setup.cfg
index 75dbef3..339f8d9 100644
--- a/setup.cfg
+++ b/setup.cfg
@@ -23,6 +23,9 @@ package_dir=
 packages=find:
 python_requires = >=3.8
 install_requires = 
+    gitpython
+    xmlschema
+    pydantic
 
 [options.packages.find]
 where=src
diff --git a/src/pynxxas/nxdl/__init__.py b/src/pynxxas/nxdl/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/src/pynxxas/nxdl/models.py b/src/pynxxas/nxdl/models.py
new file mode 100644
index 0000000..e95bab6
--- /dev/null
+++ b/src/pynxxas/nxdl/models.py
@@ -0,0 +1,329 @@
+"""NXDL models"""
+
+from typing import Optional, Union, List
+
+import pydantic
+
+
+class Item(pydantic.BaseModel):
+
+    class Config:
+        extra = "forbid"
+
+
+class EnumerationItem(Item):
+    value: str
+    doc: Optional[str] = None
+
+
+class Enumeration(Item):
+    item: List[EnumerationItem]
+
+
+class DimensionItem(Item):
+    index: int
+    required: bool
+    value: Optional[str] = None
+    ref: Optional[str] = None
+
+
+class DocDimensionItem(Item):
+    index: str
+    value: str
+
+
+class DocDimensions(Item):
+    dim: List[DocDimensionItem]
+
+
+class Dimensions(Item):
+    rank: Optional[Union[int, str]] = None
+    dim: Optional[List[DimensionItem]] = None
+    doc: Optional[Union[DocDimensions, str]] = None
+
+
+class Attribute(Item):
+    name: str
+    type: str
+    doc: Optional[str] = None
+    recommended: Optional[bool] = None
+    optional: Optional[bool] = None
+    dimensions: Optional[Dimensions] = None
+    enumeration: Optional[Enumeration] = None
+    deprecated: Optional[str] = None
+
+
+class Field(Item):
+    name: str
+    type: str
+    nameType: Optional[str] = None
+    units: Optional[str] = None
+    signal: Optional[int] = None
+    axis: Optional[int] = None
+    primary: Optional[int] = None
+    axes: Optional[str] = None
+    doc: Optional[str] = None
+    recommended: Optional[bool] = None
+    optional: Optional[bool] = None
+    minOccurs: Optional[int] = None
+    maxOccurs: Optional[Union[int, str]] = None
+    stride: Optional[bool] = None
+    data_offset: Optional[bool] = None
+    dimensions: Optional[Dimensions] = None
+    enumeration: Optional[Enumeration] = None
+    attribute: Optional[List[Attribute]] = None
+    deprecated: Optional[str] = None
+
+    class Config:
+        extra = "forbid"
+
+
+class Group(Item):
+    type: str
+    name: Optional[str] = None  # type[2:].upper()
+    doc: Optional[List[Optional[str]]] = None  # NXmirror returns [None]
+    recommended: Optional[bool] = None
+    optional: Optional[bool] = None
+    minOccurs: Optional[int] = None
+    maxOccurs: Optional[Union[int, str]] = None
+    attribute: Optional[List[Attribute]] = None
+    field: Optional[List[Field]] = None
+    group: Optional[List["Group"]] = None
+    link: Optional[List["Link"]] = None
+    deprecated: Optional[str] = None
+
+    @pydantic.model_validator(mode="after")
+    def name_from_type(self) -> "Group":
+        if self.name is None:
+            self.name = self.type[2:].upper()
+        return self
+
+
+class Link(Item):
+    name: str
+    target: str
+    doc: Optional[str] = None
+
+    class Config:
+        extra = "forbid"
+
+
+class Choice(Item):
+    name: str
+    group: List[Group]
+
+    class Config:
+        extra = "forbid"
+
+
+class Symbol(Item):
+    name: str
+    doc: str
+
+
+class Symbols(Item):
+    symbol: Optional[List[Symbol]] = None
+    doc: Optional[str] = None
+
+
+class Definition(Item):
+    xmlns: str
+    xmlns_xsi: str
+    name: str
+    type: str
+    category: str
+    xsi_schemaLocation: str
+    ignoreExtraGroups: bool
+    ignoreExtraFields: bool
+    ignoreExtraAttributes: bool
+    xmlns_xs: Optional[str] = None
+    xmlns_ns: Optional[str] = None
+    extends: Optional[str] = None
+    deprecated: Optional[str] = None
+    doc: Optional[str] = None
+    symbols: Optional[Symbols] = None
+    attribute: Optional[List[Attribute]] = None
+    field: Optional[List[Field]] = None
+    group: Optional[List[Group]] = None
+    link: Optional[List[Link]] = None
+    choice: Optional[List[Choice]] = None
+
+    class Config:
+        extra = "forbid"
+
+
+def load_enumeration_item(enum_item: dict) -> dict:
+    data = dict()
+    for key, value in enum_item.items():
+        if key.startswith("@"):
+            key = key[1:]
+        data[key] = value
+    return data
+
+
+def load_enumeration(enumeration: dict) -> dict:
+    data = dict()
+    for key, value in enumeration.items():
+        data[key] = [load_enumeration_item(item) for item in value]
+    return data
+
+
+def load_doc_dimension_item(dim_item: dict) -> dict:
+    data = dict()
+    for key, value in dim_item.items():
+        if key.startswith("@"):
+            key = key[1:]
+        data[key] = value
+    return data
+
+
+def load_dimension_item(dim_item: dict) -> dict:
+    data = dict()
+    for key, value in dim_item.items():
+        if key.startswith("@"):
+            key = key[1:]
+        data[key] = value
+    return data
+
+
+def load_doc_dimensions(dimensions: dict) -> dict:
+    data = dict()
+    for key, value in dimensions.items():
+        if key.startswith("@"):
+            key = key[1:]
+        elif key == "dim":
+            value = [load_doc_dimension_item(item) for item in value]
+        data[key] = value
+    return data
+
+
+def load_dimensions(dimensions: dict) -> dict:
+    data = dict()
+    for key, value in dimensions.items():
+        if key.startswith("@"):
+            key = key[1:]
+        elif key == "dim":
+            value = [load_dimension_item(item) for item in value]
+        elif key == "doc" and isinstance(value, dict):
+            value = load_doc_dimensions(value)
+        data[key] = value
+    return data
+
+
+def load_link(link: dict) -> dict:
+    data = dict()
+    for key, value in link.items():
+        if key.startswith("@"):
+            key = key[1:]
+        data[key] = value
+    return data
+
+
+def load_choice(choice: dict) -> dict:
+    data = dict()
+    for key, value in choice.items():
+        if key.startswith("@"):
+            key = key[1:]
+        elif key == "group":
+            value = [load_group(group) for group in value]
+        data[key] = value
+    return data
+
+
+def load_attribute(attr: dict) -> dict:
+    data = dict()
+    for key, value in attr.items():
+        if key.startswith("@"):
+            key = key[1:]
+        elif key == "enumeration":
+            value = load_enumeration(value)
+        elif key == "dimensions":
+            value = load_dimensions(value)
+        data[key] = value
+    return data
+
+
+def load_field(field: dict) -> dict:
+    data = dict()
+    for key, value in field.items():
+        if key.startswith("@"):
+            key = key[1:]
+        elif key == "attribute":
+            value = [load_attribute(attr) for attr in value]
+        elif key == "enumeration":
+            value = load_enumeration(value)
+        elif key == "dimensions":
+            value = load_dimensions(value)
+        data[key] = value
+    return data
+
+
+def load_group(group: dict) -> dict:
+    data = dict()
+    for key, value in group.items():
+        if key.startswith("@"):
+            key = key[1:]
+        elif key == "attribute":
+            value = [load_attribute(attr) for attr in value]
+        elif key == "field":
+            value = [load_field(field) for field in value]
+        elif key == "group":
+            value = [load_group(group) for group in value]
+        elif key == "link":
+            value = [load_link(attr) for attr in value]
+        data[key] = value
+    return data
+
+
+def load_symbol(symbol: dict) -> dict:
+    data = dict()
+    for key, value in symbol.items():
+        if key.startswith("@"):
+            key = key[1:]
+        data[key] = value
+    return data
+
+
+def load_symbols(symbols: dict) -> dict:
+    data = dict()
+    for key, value in symbols.items():
+        if key.startswith("@"):
+            key = key[1:]
+        elif key == "symbol":
+            value = [load_symbol(attr) for attr in value]
+        data[key] = value
+    return data
+
+
+def load_definition(definition: str) -> Definition:
+    data = dict()
+
+    for key, value in definition.items():
+        if key.startswith("@"):
+            key = key[1:].replace(":", "_")
+        elif key == "attribute":
+            value = [load_attribute(attr) for attr in value]
+        elif key == "field":
+            value = [load_field(attr) for attr in value]
+        elif key == "group":
+            value = [load_group(attr) for attr in value]
+        elif key == "link":
+            value = [load_link(attr) for attr in value]
+        elif key == "choice":
+            value = [load_choice(attr) for attr in value]
+        elif key == "symbols":
+            value = load_symbols(value)
+        data[key] = value
+
+    return Definition(**data)
+
+
+if __name__ == "__main__":
+    from . import repo
+
+    names = repo.get_nxdl_class_names(
+        url="https://github.com/XraySpectroscopy/nexus_definitions.git"
+    )
+
+    for name in names:
+        load_definition(repo.get_nxdl_class(name))
diff --git a/src/pynxxas/nxdl/repo.py b/src/pynxxas/nxdl/repo.py
new file mode 100644
index 0000000..678093f
--- /dev/null
+++ b/src/pynxxas/nxdl/repo.py
@@ -0,0 +1,75 @@
+"""NXDL repository"""
+
+import os
+import tempfile
+from glob import glob
+from functools import lru_cache
+from typing import Optional
+
+import git
+import xmlschema
+
+DEFAULT_URL = "https://github.com/nexusformat/definitions.git"
+
+
+@lru_cache(maxsize=1)
+def get_nxdl_schema(**repo_options):
+    """Returns the NDXL schema"""
+    repo = _get_repo(**repo_options)
+    schema_file = os.path.join(repo.working_dir, "nxdl.xsd")
+    return xmlschema.XMLSchema(schema_file, validation="lax")
+
+
+@lru_cache(maxsize=1)
+def get_nxdl_class_names(**repo_options):
+    """Returns all nxdl file names from the repo"""
+    working_dir = _get_repo(**repo_options).working_dir
+    pattern = os.path.join(working_dir, "*", "*.nxdl.xml")
+    return [
+        os.path.basename(filename).replace(".nxdl.xml", "")
+        for filename in glob(pattern)
+    ]
+
+
+def get_nxdl_class(name: str, **repo_options) -> dict:
+    """Load the content of an nxdl file"""
+    schema = get_nxdl_schema(**repo_options)
+    base_url = schema.base_url.replace("file://", "")
+    for dirname in ("base_classes", "applications", "contributed_definitions"):
+        xml_file = os.path.join(base_url, dirname, f"{name}.nxdl.xml")
+        if os.path.exists(xml_file):
+            break
+    return schema.to_dict(xml_file, process_namespaces=True, use_defaults=True)
+
+
+@lru_cache(maxsize=1)
+def _get_repo(
+    nxdl_version: Optional[str] = None,
+    localdir: Optional[str] = None,
+    url: Optional[str] = None,
+    branch: Optional[str] = None,
+) -> git.Repo:
+    """Git repository with NeXus definition files (*.nxdl.xml)"""
+    if not localdir:
+        localdir = os.path.join(tempfile.gettempdir(), "nexus_definitions")
+    if not url:
+        url = DEFAULT_URL
+    if not branch:
+        branch = "main"
+    remote = "origin"
+
+    if os.path.exists(localdir):
+        repo = git.Repo(localdir)
+        origin = repo.remotes[remote]
+        origin.set_url(url)
+        origin.fetch()
+        origin.pull()
+    else:
+        repo = git.Repo.clone_from(url, localdir)
+
+    if nxdl_version:
+        repo.git.checkout(f"v{nxdl_version}")
+    else:
+        repo.git.checkout(branch)
+        repo.git.reset(f"{remote}/{branch}", hard=True)
+    return repo
diff --git a/src/pynxxas/tests/conftest.py b/src/pynxxas/tests/conftest.py
new file mode 100644
index 0000000..2f4092a
--- /dev/null
+++ b/src/pynxxas/tests/conftest.py
@@ -0,0 +1,8 @@
+import pytest
+from ..nxdl import repo
+
+
+@pytest.fixture(scope="session")
+def repo_directory(tmpdir_factory) -> str:
+    root = tmpdir_factory.mktemp("nexus_definitions")
+    return repo._get_repo(localdir=str(root / "official_repo")).working_dir
diff --git a/src/pynxxas/tests/test_nxdl.py b/src/pynxxas/tests/test_nxdl.py
new file mode 100644
index 0000000..1a12315
--- /dev/null
+++ b/src/pynxxas/tests/test_nxdl.py
@@ -0,0 +1,10 @@
+from ..nxdl import repo
+from ..nxdl import models
+
+
+def test_nxdl_models(repo_directory):
+    names = repo.get_nxdl_class_names(localdir=repo_directory)
+
+    for name in names:
+        definition = repo.get_nxdl_class(name, localdir=repo_directory)
+        assert models.load_definition(definition)
diff --git a/src/pynxxas/tests/test_todo.py b/src/pynxxas/tests/test_todo.py
deleted file mode 100644
index 4f6c6c3..0000000
--- a/src/pynxxas/tests/test_todo.py
+++ /dev/null
@@ -1,2 +0,0 @@
-def test_todo():
-    pass