diff --git a/countess/plugins/collate.py b/countess/plugins/collate.py
new file mode 100644
index 0000000..a8097cc
--- /dev/null
+++ b/countess/plugins/collate.py
@@ -0,0 +1,84 @@
+from typing import Iterable, List
+
+import numpy as np
+import pandas as pd
+
+from countess import VERSION
+from countess.core.logger import Logger
+from countess.core.parameters import BooleanParam, ChoiceParam, ColumnOrNoneChoiceParam, IntegerParam, PerColumnArrayParam, TabularMultiParam
+from countess.core.plugins import PandasProcessPlugin
+from countess.utils.pandas import flatten_columns, get_all_columns
+
+
+class CollatePlugin(PandasProcessPlugin):
+    """Collates, sorts and selects data"""
+
+    name = "Collate"
+    description = "Collate and sort records by column(s), taking the first N"
+    version = VERSION
+
+    input_columns: dict[str, np.dtype]
+
+    parameters = {
+        "columns": PerColumnArrayParam(
+            "Columns",
+            ChoiceParam("Role", choices=["—", "Group", "Sort (Asc)", "Sort (Desc)"])
+        ),
+        "limit": IntegerParam("First N records", 0),
+    }
+
+    dataframes: List[pd.DataFrame]
+
+    def __init__(self, *a, **k):
+        super().__init__(*a, **k)
+        self.prepare()
+
+    def prepare(self, *_):
+        self.dataframes = []
+        self.input_columns = {}
+
+    def process(self, data: pd.DataFrame, source: str, logger: Logger) -> Iterable:
+        # XXX should do this in two stages: group each dataframe and then combine.
+        # that can wait for a more general MapReduceFinalizePlugin class though.
+        assert self.dataframes is not None
+
+        self.dataframes.append(data)
+        self.input_columns.update(get_all_columns(data))
+        return []
+
+    def sort_and_limit(self, df):
+        column_parameters = list(zip(self.input_columns, self.parameters["columns"]))
+        sort_cols = [col for col, param in column_parameters if param.value.startswith("Sort")]
+        sort_dirs = [
+            param.value.endswith("(Asc)")
+            for param in self.parameters["columns"]
+            if param.value.startswith("Sort")
+        ]
+
+        df = df.sort_values(by=sort_cols, ascending=sort_dirs)
+        if self.parameters["limit"].value > 0:
+            df = df.head(self.parameters["limit"].value)
+        return df
+
+    def finalize(self, logger: Logger) -> Iterable[pd.DataFrame]:
+        assert isinstance(self.parameters["columns"], PerColumnArrayParam)
+        assert self.dataframes
+
+        self.parameters["columns"].set_column_choices(self.input_columns.keys())
+
+        column_parameters = list(zip(self.input_columns, self.parameters["columns"]))
+        group_cols = [col for col, param in column_parameters if param.value == "Group"]
+
+        df = pd.concat(self.dataframes)
+
+        try:
+            if group_cols:
+                df = df.groupby(group_cols, group_keys=False).apply(self.sort_and_limit)
+            else:
+                df = self.sort_and_limit(df)
+
+            print(f">>>>> {type(df)}\n{df}")
+
+            yield df
+        except ValueError as exc:
+            logger.exception(exc)
diff --git a/pyproject.toml b/pyproject.toml
index a6820a5..7beecf4 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -47,6 +47,7 @@ hdf = [
 ]
 
 [project.entry-points.countess_plugins]
+collate = "countess.plugins.collate:CollatePlugin"
 load_fastq = "countess.plugins.fastq:LoadFastqPlugin"
 load_hdf = "countess.plugins.hdf5:LoadHdfPlugin"
 load_csv = "countess.plugins.csv:LoadCsvPlugin"