diff --git a/scripts/iib/topic_cluster.py b/scripts/iib/topic_cluster.py
index 1127370..e7b7250 100644
--- a/scripts/iib/topic_cluster.py
+++ b/scripts/iib/topic_cluster.py
@@ -18,16 +18,31 @@ from pydantic import BaseModel
 from scripts.iib.db.datamodel import DataBase, ImageEmbedding, ImageEmbeddingFail, TopicClusterCache, TopicTitleCache
 from scripts.iib.tool import cwd
 
-# Optional perf deps (heavy but worth it for 10k+ images)
-try:
-    import numpy as _np  # type: ignore
-except Exception:  # pragma: no cover
-    _np = None
+# Perf deps (required for this feature)
+_np = None
+_hnswlib = None
+_PERF_DEPS_READY = False
 
-try:
-    import hnswlib as _hnswlib  # type: ignore
-except Exception:  # pragma: no cover
-    _hnswlib = None
+
+def _ensure_perf_deps() -> None:
+    """
+    We do NOT allow users to use TopicSearch/TopicClustering feature without these deps.
+    But we also do NOT want the whole server to crash on import, so we lazy-load and fail at API layer.
+    """
+    global _np, _hnswlib, _PERF_DEPS_READY
+    if _PERF_DEPS_READY:
+        return
+    try:
+        import numpy as np  # type: ignore
+        import hnswlib as hnswlib  # type: ignore
+    except Exception as e:
+        raise HTTPException(
+            status_code=500,
+            detail=f"Topic clustering requires numpy+hnswlib. Please install them. import_error={type(e).__name__}: {e}",
+        )
+    _np = np
+    _hnswlib = hnswlib
+    _PERF_DEPS_READY = True
 
 
 _TOPIC_CLUSTER_JOBS: Dict[str, Dict] = {}
@@ -333,18 +348,13 @@ def _dot(a: array, b: array) -> float:
     return sum((x * y for x, y in zip(a, b)))
 
 
-def _can_use_ann() -> bool:
-    return _np is not None and _hnswlib is not None
-
-
 def _centroid_vec_np(sum_vec: array, norm_sq: float):
     """
     Convert centroid sum-vector to a normalized numpy float32 vector.
     Cosine between unit v and centroid is dot(v, sum)/sqrt(norm_sq).
     So centroid direction is sum / ||sum||.
     """
-    if _np is None:
-        return None
+    _ensure_perf_deps()
     if norm_sq <= 0:
         return None
     inv = 1.0 / math.sqrt(norm_sq)
@@ -359,7 +369,8 @@ def _build_hnsw_index(centroids_np, *, ef: int = 64, M: int = 32):
     Build a cosine HNSW index over centroid vectors.
     Returns index or None.
     """
-    if not _can_use_ann() or centroids_np is None:
+    _ensure_perf_deps()
+    if centroids_np is None:
         return None
     if len(centroids_np) == 0:
         return None
@@ -620,6 +631,9 @@ def mount_topic_cluster_routes(
     """
     Mount embedding + topic clustering endpoints (MVP: manual, iib_output only).
     """
+    # Fail fast at API layer if required perf deps are missing.
+    # (We don't crash the whole server at import time.)
+    _ensure_perf_deps()
 
     async def _run_cluster_job(job_id: str, req) -> None:
         try:
@@ -1142,7 +1156,7 @@ def mount_topic_cluster_routes(
             best_sim = -1.0
             best_dot = 0.0
             # Build / rebuild ANN index when helpful (many clusters)
-            if _can_use_ann() and len(clusters) >= 64:
+            if len(clusters) >= 64:
                 if ann_idx is None or (idx % ann_rebuild_every == 0):
                     # rebuild from current centroids
                     cents = []
@@ -1226,7 +1240,7 @@ def mount_topic_cluster_routes(
                         new_large.append(c)
                 # Build ANN over large centroids once (optional)
                 ann_large = None
-                if _can_use_ann() and len(new_large) >= 64:
+                if len(new_large) >= 64:
                     cents = []
                     for c in new_large:
                         cv = _centroid_vec_np(c["sum"], c["norm_sq"])