diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
new file mode 100644
index 0000000..851be7a
--- /dev/null
+++ b/.github/workflows/ci.yml
@@ -0,0 +1,50 @@
+name: CI
+
+on:
+  workflow_dispatch:
+  push:
+    branches: [main, master]
+  pull_request:
+    branches: [main, master]
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        python-version: ["3.12"]
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Set up Python ${{ matrix.python-version }}
+        uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python-version }}
+
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip
+          pip install pytest
+          pip install -e .
+          pip install -r requirements.txt
+
+      - name: Detect GPU availability
+        id: detect-gpu
+        shell: bash
+        run: |
+          HAS_GPU=$(python -c "import torch; print('true' if torch.cuda.is_available() else 'false')")
+          echo "has_gpu=$HAS_GPU" >> "$GITHUB_OUTPUT"
+          echo "GPU available: $HAS_GPU"
+
+      - name: Pre-cache model checkpoint
+        if: steps.detect-gpu.outputs.has_gpu == 'true'
+        run: |
+          mkdir -p ~/.cache/torch/hub/checkpoints
+          if [ ! -f ~/.cache/torch/hub/checkpoints/sharp_2572gikvuh.pt ]; then
+            curl -L -o ~/.cache/torch/hub/checkpoints/sharp_2572gikvuh.pt https://ml-site.cdn-apple.com/models/sharp/sharp_2572gikvuh.pt
+          fi
+
+      - name: Run tests
+        run: |
+          pytest
diff --git a/.gitignore b/.gitignore
index 49aa19b..8f1c51f 100644
--- a/.gitignore
+++ b/.gitignore
@@ -164,3 +164,4 @@ cython_debug/
 .DS_STORE
 *.pt
 .aider*
+.vscode/
diff --git a/tests/__init__.py b/tests/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/tests/data/example.jpg b/tests/data/example.jpg
new file mode 100644
index 0000000..37179e8
Binary files /dev/null and b/tests/data/example.jpg differ
diff --git a/tests/test_basic.py b/tests/test_basic.py
new file mode 100644
index 0000000..18ace9a
--- /dev/null
+++ b/tests/test_basic.py
@@ -0,0 +1,11 @@
+def test_import():
+    """Test that the package can be imported."""
+    import sharp
+    assert sharp is not None
+
+
+def test_cli_import():
+    """Test that CLI modules can be imported."""
+    from sharp import cli
+    assert cli is not None
+
diff --git a/tests/test_gaussians.py b/tests/test_gaussians.py
new file mode 100644
index 0000000..ce97e8a
--- /dev/null
+++ b/tests/test_gaussians.py
@@ -0,0 +1,73 @@
+import tempfile
+from pathlib import Path
+
+import numpy as np
+import torch
+
+from sharp.utils.gaussians import Gaussians3D, SceneMetaData, save_ply
+
+
+def test_gaussians3d_creation():
+    """Test creating a Gaussians3D object."""
+    num_gaussians = 100
+    gaussians = Gaussians3D(
+        mean_vectors=torch.randn(1, num_gaussians, 3),
+        singular_values=torch.rand(1, num_gaussians, 3),
+        quaternions=torch.randn(1, num_gaussians, 4),
+        colors=torch.rand(1, num_gaussians, 3),
+        opacities=torch.rand(1, num_gaussians),
+    )
+
+    assert gaussians.mean_vectors.shape == (1, num_gaussians, 3)
+    assert gaussians.singular_values.shape == (1, num_gaussians, 3)
+    assert gaussians.quaternions.shape == (1, num_gaussians, 4)
+    assert gaussians.colors.shape == (1, num_gaussians, 3)
+    assert gaussians.opacities.shape == (1, num_gaussians)
+
+
+def test_gaussians3d_to_device():
+    """Test moving Gaussians3D to a device."""
+    num_gaussians = 50
+    gaussians = Gaussians3D(
+        mean_vectors=torch.randn(1, num_gaussians, 3),
+        singular_values=torch.rand(1, num_gaussians, 3),
+        quaternions=torch.randn(1, num_gaussians, 4),
+        colors=torch.rand(1, num_gaussians, 3),
+        opacities=torch.rand(1, num_gaussians),
+    )
+
+    gaussians_cpu = gaussians.to(torch.device("cpu"))
+    assert gaussians_cpu.mean_vectors.device.type == "cpu"
+    assert gaussians_cpu.singular_values.device.type == "cpu"
+
+
+def test_scene_metadata():
+    """Test creating SceneMetaData."""
+    metadata = SceneMetaData(
+        focal_length_px=1000.0,
+        resolution_px=(1920, 1080),
+        color_space="linearRGB",
+    )
+
+    assert metadata.focal_length_px == 1000.0
+    assert metadata.resolution_px == (1920, 1080)
+    assert metadata.color_space == "linearRGB"
+
+
+def test_save_ply(tmp_path):
+    """Test saving Gaussians to PLY file."""
+    num_gaussians = 10
+    gaussians = Gaussians3D(
+        mean_vectors=torch.randn(1, num_gaussians, 3),
+        singular_values=torch.rand(1, num_gaussians, 3),
+        quaternions=torch.randn(1, num_gaussians, 4),
+        colors=torch.rand(1, num_gaussians, 3),
+        opacities=torch.rand(1, num_gaussians),
+    )
+
+    output_path = tmp_path / "test_gaussians.ply"
+    save_ply(gaussians, f_px=1000.0, image_shape=(1920, 1080), path=output_path)
+
+    assert output_path.exists()
+    assert output_path.stat().st_size > 0
+
diff --git a/tests/test_io.py b/tests/test_io.py
new file mode 100644
index 0000000..1b14f54
--- /dev/null
+++ b/tests/test_io.py
@@ -0,0 +1,69 @@
+import tempfile
+from pathlib import Path
+
+import numpy as np
+import pytest
+from PIL import Image
+
+from sharp.utils import io
+
+
+def test_load_rgb(tmp_path):
+    """Test loading an RGB image."""
+    test_image_path = tmp_path / "test.jpg"
+    img_array = np.random.randint(0, 255, (100, 100, 3), dtype=np.uint8)
+    img = Image.fromarray(img_array)
+    img.save(test_image_path)
+
+    image, icc_profile, f_px = io.load_rgb(test_image_path)
+
+    assert image.shape == (100, 100, 3)
+    assert image.dtype == np.uint8
+    assert f_px > 0
+    assert isinstance(icc_profile, (list, type(None)))
+
+
+def test_get_supported_image_extensions():
+    """Test getting supported image extensions."""
+    extensions = io.get_supported_image_extensions()
+    assert isinstance(extensions, list)
+    assert len(extensions) > 0
+    assert ".jpg" in extensions or ".JPG" in extensions
+    assert ".png" in extensions or ".PNG" in extensions
+
+
+def test_get_supported_video_extensions():
+    """Test getting supported video extensions."""
+    extensions = io.get_supported_video_extensions()
+    assert isinstance(extensions, list)
+    assert ".mp4" in extensions or ".MP4" in extensions
+
+
+def test_save_image(tmp_path):
+    """Test saving an image."""
+    test_image_path = tmp_path / "test_output.jpg"
+    img_array = np.random.randint(0, 255, (50, 50, 3), dtype=np.uint8)
+
+    io.save_image(img_array, test_image_path)
+
+    assert test_image_path.exists()
+    loaded_image, _, _ = io.load_rgb(test_image_path)
+    assert loaded_image.shape == (50, 50, 3)
+
+
+def test_convert_focallength():
+    """Test focal length conversion."""
+    f_px = io.convert_focallength(1920, 1080, 30.0)
+    assert f_px > 0
+    assert isinstance(f_px, float)
+
+
+def test_load_example_image():
+    """Test loading the example image from tests/data directory."""
+    example_path = Path(__file__).parent / "data" / "example.jpg"
+    if example_path.exists():
+        image, _, f_px = io.load_rgb(example_path)
+        assert image.ndim == 3
+        assert image.shape[2] == 3
+        assert f_px > 0
+
diff --git a/tests/test_predict.py b/tests/test_predict.py
new file mode 100644
index 0000000..236410b
--- /dev/null
+++ b/tests/test_predict.py
@@ -0,0 +1,75 @@
+import numpy as np
+import pytest
+import torch
+from pathlib import Path
+
+from sharp.cli.predict import predict_image
+from sharp.models import PredictorParams, create_predictor
+from sharp.utils.gaussians import Gaussians3D
+
+
+@pytest.mark.skipif(
+    not torch.cuda.is_available() and not torch.mps.is_available(),
+    reason="Requires CUDA or MPS for model inference",
+)
+def test_predict_image_with_model(tmp_path):
+    """Test predict_image function with a real model checkpoint."""
+    example_path = Path(__file__).parent / "data" / "example.jpg"
+    if not example_path.exists():
+        pytest.skip("Test image not found")
+
+    from sharp.utils import io
+
+    image, _, f_px = io.load_rgb(example_path)
+
+    device = torch.device("cuda" if torch.cuda.is_available() else "mps" if torch.mps.is_available() else "cpu")
+
+    # Use the pre-cached model checkpoint
+    checkpoint_path = Path.home() / ".cache" / "torch" / "hub" / "checkpoints" / "sharp_2572gikvuh.pt"
+    if not checkpoint_path.exists():
+        pytest.skip("Model checkpoint not found in cache")
+
+    try:
+        predictor = create_predictor(PredictorParams(checkpoint_path=str(checkpoint_path)))
+        predictor.eval()
+        predictor.to(device)
+
+        gaussians = predict_image(predictor, image, f_px, device)
+
+        assert isinstance(gaussians, Gaussians3D)
+        assert gaussians.mean_vectors.shape[0] == 1
+        assert gaussians.mean_vectors.shape[2] == 3
+        assert gaussians.colors.shape[2] == 3
+        assert gaussians.opacities.shape[1] > 0
+    except Exception as e:
+        pytest.skip(f"Model inference failed (likely missing checkpoint): {e}")
+
+
+def test_predict_image_signature():
+    """Test that predict_image function has correct signature."""
+    import inspect
+
+    sig = inspect.signature(predict_image)
+    params = sig.parameters
+
+    assert "predictor" in params
+    assert "image" in params
+    assert "f_px" in params
+    assert "device" in params
+
+
+def test_create_predictor():
+    """Test creating a predictor model."""
+    params = PredictorParams()
+    predictor = create_predictor(params)
+
+    assert predictor is not None
+    assert hasattr(predictor, "eval")
+    assert hasattr(predictor, "to")
+
+
+def test_predictor_params():
+    """Test PredictorParams creation."""
+    params = PredictorParams()
+    assert params is not None
+