diff --git a/ai_mouse/server/routes_train.py b/ai_mouse/server/routes_train.py
index 14a66e1..2c4d511 100644
--- a/ai_mouse/server/routes_train.py
+++ b/ai_mouse/server/routes_train.py
@@ -23,13 +23,17 @@ router = APIRouter()
 # ---------------------------------------------------------------------------
 
 
-def _paths() -> tuple[Path, Path]:
+def _paths() -> tuple[Path, Path, Path]:
     data_dir = get_data_dir()
-    return data_dir / "traces.jsonl", data_dir / "models_v2"
+    return (
+        data_dir / "traces.jsonl",
+        data_dir / "models_v2",
+        data_dir / "models_v2_pretrained",
+    )
 
 
 def _trace_count() -> int:
-    traces_path, _ = _paths()
+    traces_path, _, _ = _paths()
     if not traces_path.exists():
         return 0
     return sum(
@@ -40,7 +44,7 @@ def _trace_count() -> int:
 
 
 def _model_trained() -> bool:
-    _, models_dir = _paths()
+    _, models_dir, _ = _paths()
     return (models_dir / "flow_model.pt").exists()
 
 
@@ -75,16 +79,31 @@ async def _train_sse_generator(req: TrainRequest) -> AsyncGenerator[str, None]:
     async def run_training_async() -> None:
         from ai_mouse.trainer import train
 
-        traces_path, models_dir = _paths()
+        traces_path, models_dir, pretrained_dir = _paths()
         data_path = Path(req.data_path) if req.data_path else traces_path
         output_dir = Path(req.output_dir) if req.output_dir else models_dir
+
+        # Auto-detect pretrained checkpoint and switch to fine-tune mode
+        resume_from: Path | None = None
+        effective_lr = 3e-4
+        if (pretrained_dir / "flow_model.pt").exists():
+            resume_from = pretrained_dir
+            effective_lr = 1e-5  # fine-tune lr
+            logger.info("Detected pretrained checkpoint, fine-tuning at lr=%g", effective_lr)
+            queue.put_nowait({
+                "info": f"Detected pretrained checkpoint at {pretrained_dir.name}, "
+                        f"running fine-tune at lr={effective_lr}",
+            })
+
         try:
             await asyncio.to_thread(
                 train,
                 data_path=data_path,
                 output_dir=output_dir,
                 epochs=req.epochs,
+                lr=effective_lr,
                 progress_callback=callback,
+                resume_from=resume_from,
             )
         except Exception as exc:  # noqa: BLE001
             queue.put_nowait({"error": str(exc)})