diff --git a/graph_net/agent/parallel_extract.py b/graph_net/agent/parallel_extract.py
index 2c6b9fc93..42eea8bd0 100644
--- a/graph_net/agent/parallel_extract.py
+++ b/graph_net/agent/parallel_extract.py
@@ -383,13 +383,13 @@ def _parse_args() -> argparse.Namespace:
         "--gpus",
         type=str,
         default=None,
-        help="Comma-separated GPU indices to use (GPU mode; if set, ignores --num-workers)",
+        help="Comma-separated GPU indices to use (default: auto-detect all available GPUs)",
     )
     parser.add_argument(
-        "--num-workers",
+        "--cpu-workers",
         type=int,
         default=None,
-        help="Number of worker processes in CPU mode (default: CPU count)",
+        help="Number of worker processes in CPU-only mode (default: half of CPU cores)",
     )
     parser.add_argument(
         "--output",
@@ -435,22 +435,29 @@ def _resolve_config(args: argparse.Namespace):
     )
     print(f"[INFO] Workspace: {workspace}")
 
-    if get_device_type() == "cuda":
+    # Decide GPU vs CPU mode: if --cpu-workers is set, force CPU-only mode.
+    # If no CUDA available, also fall back to CPU mode.
+    if (args.cpu_workers and args.cpu_workers > 0) or get_device_type() == "cpu":
+        # CPU-only mode. Default to half of CPU cores to avoid overloading the
+        # system, since each worker is a heavy process (model loading + graph
+        # extraction).
+        gpus = []
+        num_workers = (
+            args.cpu_workers if args.cpu_workers else max(1, (os.cpu_count() or 2) // 2)
+        )
+        print(f"[INFO] CPU-only mode: {num_workers} workers")
+        extract_timeout = (
+            args.extract_timeout if args.extract_timeout is not None else 2000
+        )
+        verify_timeout = args.verify_timeout if args.verify_timeout is not None else 600
+    else:
         gpus = get_gpu_ids(args)
         num_workers = len(gpus)
-        print(f"[INFO] GPU mode (torch fallback): {gpus}")
+        print(f"[INFO] GPU mode: {num_workers} workers on GPUs {gpus}")
         extract_timeout = (
             args.extract_timeout if args.extract_timeout is not None else 1000
         )
         verify_timeout = args.verify_timeout if args.verify_timeout is not None else 300
-    else:
-        gpus = []
-        num_workers = args.num_workers if args.num_workers else 1
-        print(f"[INFO] CPU mode: {num_workers} workers")
-        extract_timeout = (
-            args.extract_timeout if args.extract_timeout is not None else 2000
-        )
-        verify_timeout = args.verify_timeout if args.verify_timeout is not None else 600
 
     return workspace, gpus, num_workers, extract_timeout, verify_timeout
 
@@ -468,7 +475,10 @@ def main() -> int:
         print("[ERROR] Empty model list, nothing to do")
         return 1
 
-    print(f"[INFO] Total models: {len(model_ids)}, workers: {num_workers}")
+    if gpus:
+        print(f"[INFO] Total models: {len(model_ids)}, GPU workers: {num_workers}")
+    else:
+        print(f"[INFO] Total models: {len(model_ids)}, CPU workers: {num_workers}")
 
     # --- Populate shared task queue ---
     task_queue: multiprocessing.Queue = multiprocessing.Queue()
@@ -479,8 +489,9 @@ def main() -> int:
     result_queue: multiprocessing.Queue = multiprocessing.Queue()
 
     start_time = datetime.now()
+    worker_type = "GPU" if gpus else "CPU"
     print(
-        f"\n[START] {start_time.strftime('%Y-%m-%d %H:%M:%S')} — launching {num_workers} workers\n"
+        f"\n[START] {start_time.strftime('%Y-%m-%d %H:%M:%S')} — launching {num_workers} {worker_type} workers\n"
     )
 
     processes = []