diff --git a/homelab-ai-bot/tools/rag.py b/homelab-ai-bot/tools/rag.py
index 0084962e..40b0ae58 100644
--- a/homelab-ai-bot/tools/rag.py
+++ b/homelab-ai-bot/tools/rag.py
@@ -22,8 +22,9 @@ EMBED_MODEL = "nomic-embed-text"
 
 # Cross-Encoder Reranking (CT 123, pve-hetzner LAN)
 RERANKER_URL = "http://10.10.10.123:8099"
-RERANK_CANDIDATES = 30
-RERANK_TIMEOUT = 120
+RERANK_CANDIDATES = 15
+RERANK_TIMEOUT = 30
+RERANK_SNIPPET_CHARS = 512
 
 MIN_TOP_K = 5
 # Breite Übersichten: mehr ES-Runden, mehr distinct Treffer (pro vollem Pfad docnm_kwd)
@@ -207,12 +208,14 @@ def _es_hybrid_search(query: str, es_size: int) -> dict:
 
 
 def _snippet_for_rerank(src: dict) -> str:
+    doc_name = src.get("docnm_kwd") or ""
     raw = src.get("content_with_weight") or src.get("content_de") or ""
-    return raw[:4000]
+    prefix = doc_name[:120] + "\n" if doc_name else ""
+    return prefix + raw[:RERANK_SNIPPET_CHARS]
 
 
 def _rerank_hits(query: str, hits: list) -> tuple[list, bool]:
-    """Sortiert die ersten RERANK_CANDIDATES Treffer per Cross-Encoder neu."""
+    """Rerankt mit Cross-Encoder, kombiniert Score mit ES-Rang (RRF)."""
     if not hits or not RERANKER_URL:
         return hits, False
     to_score = hits[:RERANK_CANDIDATES]
@@ -241,17 +244,27 @@ def _rerank_hits(query: str, hits: list) -> tuple[list, bool]:
                 len(to_score),
             )
             return hits, False
-        indexed = list(zip(scores, range(len(to_score))))
-        indexed.sort(key=lambda x: x[0], reverse=True)
+
+        k = 60
+        combined: list[tuple[float, int]] = []
+        for idx, (h, rr_score) in enumerate(zip(to_score, scores)):
+            es_rank = idx + 1
+            rr_sorted = sorted(scores, reverse=True)
+            rr_rank = rr_sorted.index(rr_score) + 1
+            rrf = 1.0 / (k + es_rank) + 1.0 / (k + rr_rank)
+            combined.append((rrf, idx))
+        combined.sort(key=lambda x: x[0], reverse=True)
+
         new_order: list = []
-        for sc, idx in indexed:
+        for rrf, idx in combined:
             h = to_score[idx]
-            h["_rerank_score"] = float(sc)
+            h["_rerank_score"] = float(scores[idx])
+            h["_rrf_score"] = float(rrf)
             new_order.append(h)
         rest = hits[RERANK_CANDIDATES:]
         return new_order + rest, True
     except Exception as e:
-        log.warning("rerank failed: %s", e)
+        log.warning("rerank failed (fallback to ES): %s", e)
         return hits, False
 
 
@@ -435,7 +448,10 @@ def handle_rag_search(query: str, top_k: int = 8, **kw):
             continue
         seen_docs.add(dk)
 
-        if "_rerank_score" in h:
+        if "_rrf_score" in h:
+            score = float(h["_rrf_score"])
+            score_label = "RRF"
+        elif "_rerank_score" in h:
             score = float(h["_rerank_score"])
             score_label = "Rerank"
         else: