Update api/server.py

api/server.py

@@ -180,6 +180,50 @@ def _should_force_rag(message: str) -> bool:
     ]
     return any(t in m for t in triggers)
 
+def _extract_filename_hint(message: str) -> Optional[str]:
+    m = (message or "").strip()
+    if not m:
+        return None
+    # 极简：如果用户直接提到了 .pdf/.ppt/.docx 文件名，就用它
+    for token in m.replace("“", '"').replace("”", '"').split():
+        if any(token.lower().endswith(ext) for ext in [".pdf", ".ppt", ".pptx", ".doc", ".docx"]):
+            return os.path.basename(token.strip('"').strip("'").strip())
+    return None
+
+
+def _resolve_rag_scope(sess: Dict[str, Any], msg: str) -> Tuple[Optional[List[str]], Optional[List[str]]]:
+    """
+    Return (allowed_source_files, allowed_doc_types)
+    - If user is asking about "uploaded file"/document action -> restrict to latest uploaded file.
+    - If message contains an explicit filename -> restrict to that filename if we have it.
+    - Else no restriction (None, None).
+    """
+    files = sess.get("uploaded_files") or []
+    msg_l = (msg or "").lower()
+
+    # 1) explicit filename mentioned
+    hinted = _extract_filename_hint(msg)
+    if hinted:
+        # only restrict if that file exists in session uploads
+        known = {os.path.basename(f.get("filename", "")) for f in files if f.get("filename")}
+        if hinted in known:
+            return ([hinted], None)
+
+    # 2) generic "uploaded file" intent
+    uploaded_intent = any(t in msg_l for t in [
+        "uploaded file", "uploaded files", "the uploaded file", "this file", "this document",
+        "上传的文件", "这份文件", "这个文件", "文档", "课件", "讲义"
+    ])
+    if uploaded_intent and files:
+        last = files[-1]
+        fn = os.path.basename(last.get("filename", "")).strip() or None
+        dt = (last.get("doc_type") or "").strip() or None
+        allowed_files = [fn] if fn else None
+        allowed_doc_types = [dt] if dt else None
+        return (allowed_files, allowed_doc_types)
+
+    return (None, None)
+
 
 # ----------------------------
 # Warmup
@@ -466,11 +510,18 @@ def chat(req: ChatReq):
 
     # NEW: do NOT bypass RAG for document actions (so UI refs are preserved)
     force_rag = _should_force_rag(msg)
-    
+
+    allowed_files, allowed_doc_types = _resolve_rag_scope(sess, 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"])
+        rag_context_text, rag_used_chunks = retrieve_relevant_chunks(
+            msg,
+            sess["rag_chunks"],
+            allowed_source_files=allowed_files,
+            allowed_doc_types=allowed_doc_types,
+        )