Update api/server.py

api/server.py

@@ -167,6 +167,19 @@ def _build_upload_hint(sess: Dict[str, Any]) -> str:
     )
     return "\n".join(lines)
 
+#  NEW: force RAG on short "document actions" so refs exist
+def _should_force_rag(message: str) -> bool:
+    m = (message or "").lower()
+    if not m:
+        return False
+    triggers = [
+        "summarize", "summary", "read", "analyze", "explain",
+        "the uploaded file", "uploaded", "file", "document", "pdf",
+        "slides", "ppt", "syllabus", "lecture",
+        "总结", "概括", "阅读", "读一下", "解析", "分析", "这份文件", "上传", "文档", "课件", "讲义",
+    ]
+    return any(t in m for t in triggers)
+
 
 # ----------------------------
 # Warmup
@@ -451,13 +464,20 @@ def chat(req: ChatReq):
     sess["cognitive_state"] = update_cognitive_state_from_message(msg, sess["cognitive_state"])
     marks_ms["cognitive_update_done"] = (time.time() - t0) * 1000.0
 
-    if len(msg) < 20 and ("?" not in msg):
+    # NEW: do NOT bypass RAG for document actions (so UI refs are preserved)
+    force_rag = _should_force_rag(msg)
+    
+    if (len(msg) < 20 and ("?" not in msg)) and (not force_rag):
         rag_context_text, rag_used_chunks = "", []
     else:
         rag_context_text, rag_used_chunks = retrieve_relevant_chunks(msg, sess["rag_chunks"])
+
+
+
+        
     marks_ms["rag_retrieve_done"] = (time.time() - t0) * 1000.0
 
-    # ✅ NEW: prepend deterministic upload/file-state hint so the model never says “no file”
+    #  NEW: prepend deterministic upload/file-state hint so the model never says “no file”
     upload_hint = _build_upload_hint(sess)
     if upload_hint:
         rag_context_text = (upload_hint + "\n\n---\n\n" + (rag_context_text or "")).strip()