diff --git a/studio/backend/models/datasets.py b/studio/backend/models/datasets.py
index 81c2451ebe..d714cd397d 100644
--- a/studio/backend/models/datasets.py
+++ b/studio/backend/models/datasets.py
@@ -21,3 +21,5 @@ class CheckFormatResponse(BaseModel):
     suggested_mapping: Optional[Dict[str, str]] = None
     detected_image_column: Optional[str] = None
     detected_text_column: Optional[str] = None
+    preview_samples: Optional[List[Dict]] = None
+    total_rows: Optional[int] = None
diff --git a/studio/backend/routes/datasets.py b/studio/backend/routes/datasets.py
index 8e21acaa9e..d6275fe742 100644
--- a/studio/backend/routes/datasets.py
+++ b/studio/backend/routes/datasets.py
@@ -37,12 +37,15 @@ async def check_format(request: CheckFormatRequest):
     """
     Check if a dataset requires manual column mapping.
     
-    This is a lightweight check that only runs format detection,
-    not full processing. Use before starting training to determine
-    if the user needs to manually map columns.
+    This is a lightweight check that loads only the first 10 rows,
+    runs format detection, and (if processable) returns processed
+    preview samples. The full dataset is re-processed at training time.
     """
     try:
         from datasets import load_dataset
+        from utils.datasets import format_dataset
+        
+        PREVIEW_SIZE = 10
         
         logger.info(f"Checking format for dataset: {request.dataset_name}")
         
@@ -69,11 +72,29 @@ async def check_format(request: CheckFormatRequest):
                 load_kwargs["token"] = request.hf_token
             dataset = load_dataset(**load_kwargs)
         
-        # Run lightweight format check
-        result = check_dataset_format(dataset, is_vlm=request.is_vlm)
+        # Slice to top N rows — all detection and preview runs on this subset
+        total_rows = len(dataset)
+        preview_slice = dataset.select(range(min(PREVIEW_SIZE, total_rows)))
+        
+        # Run lightweight format check on the preview slice
+        result = check_dataset_format(preview_slice, is_vlm=request.is_vlm)
         
         logger.info(f"Format check result: requires_mapping={result['requires_manual_mapping']}, format={result['detected_format']}")
         
+        # If format is processable, generate preview samples via format_dataset
+        preview_samples = None
+        if not result["requires_manual_mapping"]:
+            try:
+                format_result = format_dataset(
+                    preview_slice,
+                    format_type="auto",
+                    custom_format_mapping=result.get("suggested_mapping"),
+                )
+                processed = format_result["dataset"]
+                preview_samples = [dict(row) for row in processed]
+            except Exception as e:
+                logger.warning(f"Preview generation failed (non-fatal): {e}")
+        
         return CheckFormatResponse(
             requires_manual_mapping=result["requires_manual_mapping"],
             detected_format=result["detected_format"],
@@ -81,6 +102,8 @@ async def check_format(request: CheckFormatRequest):
             suggested_mapping=result.get("suggested_mapping"),
             detected_image_column=result.get("detected_image_column"),
             detected_text_column=result.get("detected_text_column"),
+            preview_samples=preview_samples,
+            total_rows=total_rows,
         )
         
     except HTTPException:
diff --git a/studio/backend/run.py b/studio/backend/run.py
index 6f93cba1d3..8945d438bb 100644
--- a/studio/backend/run.py
+++ b/studio/backend/run.py
@@ -14,7 +14,7 @@ if str(backend_dir) not in sys.path:
 def run_server(
     host: str = "0.0.0.0",
     port: int = 8000,
-    frontend_path: Path = None,
+    frontend_path: Path = "../frontend/dist",
     silent: bool = False,
 ):
     """
@@ -75,7 +75,7 @@ if __name__ == "__main__":
     parser.add_argument("--host", default="0.0.0.0", help="Host to bind to")
     parser.add_argument("--port", type=int, default=8000, help="Port to bind to")
     parser.add_argument(
-        "--frontend", type=str, default=None, help="Path to frontend build"
+        "--frontend", type=str, default="../frontend/dist", help="Path to frontend build"
     )
     parser.add_argument("--silent", action="store_true", help="Suppress output")
 
diff --git a/studio/frontend/src/features/chat/adapter.ts b/studio/frontend/src/features/chat/adapter.ts
index 08e4da9c6d..1c731cecaa 100644
--- a/studio/frontend/src/features/chat/adapter.ts
+++ b/studio/frontend/src/features/chat/adapter.ts
@@ -11,7 +11,7 @@ function collectTextParts(message: RunMessage): string[] {
     .map((c) => c.text);
 
   if ("attachments" in message && (message.attachments?.length ?? 0) > 0) {
-    for (const att of message.attachments) {
+    for (const att of message.attachments ?? []) {
       for (const part of att.content ?? []) {
         if (part.type === "text") {
           textParts.push(part.text);