optional context

app.py

@@ -79,11 +79,15 @@ def add_docs(user_id: str, docs: list[str]) -> int:
 def build_qwen_prompt(context: str, user_question: str) -> str:
     """Return a string that follows Qwen-Chat’s template."""
     load_chat()                       # ← make sure tokenizer is ready
+    if context:
+        context = f"Context:\n{context}\n\n"
+    else:
+        context = ""
     conversation = [
         {"role": "system",
          "content": "You are an email assistant. Use ONLY the context provided."},
         {"role": "user",
-         "content": f"Context:\n{context}\n\n{user_question}"}
+         "content": f"{context}{user_question}"}
     ]
     return tokenizer.apply_chat_template(
         conversation, tokenize=False, add_generation_prompt=True
@@ -98,26 +102,28 @@ def store_doc(doc_text: str, user_id="demo"):
         return "Nothing stored (empty input)."
     return f"Stored — KB now has {len(kb[user_id]['texts'])} passage(s)."
 
-def answer(question: str, user_id="demo"):
+def answer(question: str, user_id="demo", history=False):
     """UI callback: retrieve, build prompt with Qwen tags, generate answer."""
     try:
         if not question.strip():
             return "Please ask a question."
-        if not kb[user_id]["texts"]:
+        if history and not kb[user_id]["texts"]:
             return "No reference passage yet. Add one first."
 
-        # 1️⃣  Retrieve top-k similar passages
-        q_vec  = embed(question)
-        store  = kb[user_id]
-        sims   = torch.matmul(store["vecs"], q_vec)          # [N]
-        k      = min(4, sims.numel())
-        idxs   = torch.topk(sims, k=k).indices.tolist()
-        context = "\n".join(store["texts"][i] for i in idxs)
-
-        # 2️⃣  Build a Qwen-chat prompt (helper defined earlier)
+        context = None
+        # 1.  Retrieve top-k similar passages
+        if history:
+            q_vec  = embed(question)
+            store  = kb[user_id]
+            sims   = torch.matmul(store["vecs"], q_vec)          # [N]
+            k      = min(4, sims.numel())
+            idxs   = torch.topk(sims, k=k).indices.tolist()
+            context = "\n".join(store["texts"][i] for i in idxs)
+
+        # 2.  Build a Qwen-chat prompt (helper defined earlier)
         prompt = build_qwen_prompt(context, question)
 
-        # 3️⃣  Generate and strip everything before the assistant tag
+        # 3.  Generate and strip everything before the assistant tag
         load_chat()
         inputs = tokenizer(prompt, return_tensors="pt").to(chat_model.device)
         output = chat_model.generate(**inputs, max_new_tokens=512)