Add fix for generating embeddings

src/trowel/cli.py

@@ -340,6 +340,7 @@ def generate_embeddings(input_file, collection, db_path, text_fields, limit, ski
                     db_path,
                     collection,
                     export_csv,
+                    include_embeddings=True,
                 )
                 logging.info(f"Exported {num_exported} embeddings to {export_csv}")
                 logging.info("You can now use these embeddings with other embedding commands:")

src/trowel/utils/embedding_generation_utils.py

@@ -212,7 +212,11 @@ def export_embeddings_to_csv(
         # Get field names for the collection. Some CurateGPT backends can return
         # empty field_names even when rows exist, so fall back to inferring them.
         field_names = store.field_names(collection=collection_name) or []
-        raw_results = list(store.find(where={}, collection=collection_name))
+        find_kwargs = {"where": {}, "collection": collection_name}
+        if include_embeddings:
+            find_kwargs["include"] = ["metadatas", "documents", "embeddings"]
+
+        raw_results = list(store.find(**find_kwargs))
         docs = []
         for result in raw_results:
             normalized = _normalize_store_result(

tests/test_cli_embeddings.py

@@ -132,7 +132,12 @@ def test_command_with_export(self, mock_generate, mock_export, runner, sample_cs
             assert result.exit_code == 0
             assert "Exporting embeddings to" in result.output
             assert "Exported 3 embeddings" in result.output
-            assert mock_export.called
+            mock_export.assert_called_once_with(
+                db_path,
+                "embeddings",
+                export_path,
+                include_embeddings=True,
+            )
 
     @patch("trowel.cli.generate_embeddings_with_curategpt")
     def test_command_with_limit(self, mock_generate, runner, sample_csv, temp_dir):

tests/test_embedding_generation.py

@@ -422,3 +422,49 @@ def test_csv_export_handles_duckdb_tuple_results(self, mock_get_store, temp_dir)
             rows = list(reader)
             assert len(rows) == 2
             assert rows[0]["id"] == "BERVO:0000001"
+
+    @patch("trowel.utils.embedding_generation_utils._get_curategpt_store")
+    def test_csv_export_requests_duckdb_embeddings_when_included(
+        self,
+        mock_get_store,
+        temp_dir,
+    ):
+        """Test export requests embeddings when requested."""
+        db_path = os.path.join(temp_dir, "test.duckdb")
+        os.makedirs(db_path, exist_ok=True)
+
+        mock_store = MagicMock()
+        mock_get_store.return_value = mock_store
+        mock_store.field_names.return_value = []
+
+        def fake_find(where, collection, include=None):
+            embeddings = (
+                [0.1, 0.2]
+                if include and "embeddings" in include
+                else None
+            )
+            return [
+                (
+                    {"id": "BERVO:0000001", "label": "Temperature"},
+                    0.0,
+                    {"_embeddings": embeddings, "documents": "Temperature"},
+                )
+            ]
+
+        mock_store.find.side_effect = fake_find
+
+        output_path = os.path.join(temp_dir, "output.csv")
+        num_exported = export_embeddings_to_csv(
+            db_path,
+            "test_collection",
+            output_path,
+            include_embeddings=True,
+        )
+
+        assert num_exported == 1
+        with open(output_path, 'r') as f:
+            reader = csv.DictReader(f)
+            rows = list(reader)
+            assert len(rows) == 1
+            assert "embeddings" in reader.fieldnames
+            assert rows[0]["embeddings"] == "[0.1, 0.2]"