diff --git a/src/eigen_tour/__init__.py b/src/eigen_tour/__init__.py
index 100d49d..fb9a0d2 100644
--- a/src/eigen_tour/__init__.py
+++ b/src/eigen_tour/__init__.py
@@ -1,8 +1,11 @@
+from io import BytesIO
 import importlib.metadata
 import pathlib
 
 import anywidget
 import traitlets
+import pandas as pd
+import pyarrow as pa
 
 try:
     __version__ = importlib.metadata.version("eigen_tour")
@@ -10,7 +13,50 @@
     __version__ = "unknown"
 
 
+def arrow_from_pandas(df):
+    table = pa.Table.from_pandas(df)
+    with BytesIO() as stream:
+        writer = pa.ipc.new_file(stream, table.schema)
+        writer.write_table(table)
+        writer.close()
+        return stream.getvalue()
+
+
 class Widget(anywidget.AnyWidget):
     _esm = pathlib.Path(__file__).parent / "static" / "widget.js"
     _css = pathlib.Path(__file__).parent / "static" / "widget.css"
     data = traitlets.Bytes().tag(sync=True)
+    axis_fields = traitlets.List().tag(sync=True)
+    label_field = traitlets.Unicode().tag(sync=True)
+    label_colors = traitlets.Dict().tag(sync=True)
+
+    def __init__(self, df, axis_fields, label_field, label_colors, **kwargs):
+
+        if not all(field in df.columns for field in axis_fields):
+            raise ValueError("`axis_fields` must be a subset of df columns")
+
+        if label_field not in df.columns:
+            raise ValueError("`label_field` must be a column in df")
+
+        if not isinstance(df[label_field].dtype, pd.CategoricalDtype):
+            df[label_field] = df[label_field].astype("category")
+
+        label_cat = df[label_field].cat.categories
+        if isinstance(label_colors, dict):
+            for cat in label_cat:
+                if cat not in label_colors:
+                    raise ValueError(
+                        "`label_field` categories must be keys in `label_colors` dict"
+                    )
+        elif isinstance(label_colors, list):
+            label_colors = dict(zip(label_cat, label_colors))
+        else:
+            raise ValueError("`label_colors` must be a list or a dict")
+
+        super().__init__(
+            data=arrow_from_pandas(df[[label_field, *axis_fields]]),
+            axis_fields=axis_fields,
+            label_field=label_field,
+            label_colors=label_colors,
+            **kwargs,
+        )