Update app.py via AI Editor

app.py

@@ -236,7 +236,7 @@ def chat_box_card():
         dbc.CardBody([
             html.Div(id="chat-window", style={
                 "height": "60vh",
-                "overflowY": "auto",
+                "overflowY": "scroll",
                 "overflowX": "auto",
                 "display": "flex",
                 "flexDirection": "column",
@@ -247,7 +247,11 @@ def chat_box_card():
             })
         ]),
         className="mt-3",
-        style={"height": "62vh", "overflowY": "auto", "overflowX": "auto"}
+        style={
+            "height": "62vh",
+            "overflowY": "scroll",
+            "overflowX": "auto"
+        }
     )
 
 def user_input_card():
@@ -429,6 +433,7 @@ def main_callback(session_id, send_clicks, file_contents, new_chat_clicks, strea
 
         file_was_uploaded_and_sent = False
         file_upload_message = None
+        doc_texts_to_send = []
         if trigger == "file-upload" and file_contents and file_names:
             uploads = []
             file_upload_messages = []
@@ -450,12 +455,12 @@ def main_callback(session_id, send_clicks, file_contents, new_chat_clicks, strea
                     if text.strip():
                         embed_user_doc(session_id, fname, text)
                         logger.info(f"Session {session_id}: Uploaded doc '{n}' embedded for user vector store")
-                        # Add extracted text to chat
                         preview = text[:1000]
                         file_upload_messages.append({
                             "role": "user",
                             "content": f"[Document uploaded: {n}]\n{preview if preview.strip() else '[No text extracted]'}"
                         })
+                        doc_texts_to_send.append(text.strip())
                     else:
                         file_upload_messages.append({
                             "role": "user",
@@ -472,11 +477,88 @@ def main_callback(session_id, send_clicks, file_contents, new_chat_clicks, strea
                         "content": f"[File uploaded: {n}]"
                     })
             state["uploads"].extend(uploads)
-            # Only show the latest file upload message(s)
+            # Add all file upload messages to chat
             for msg in file_upload_messages:
                 state["messages"].append(msg)
             save_session_state(session_id)
             logger.info(f"Session {session_id}: Uploaded files {[u['name'] for u in uploads]}")
+            # If text was extracted, send it to OpenAI as a question for assistant reply
+            if doc_texts_to_send:
+                doc_question = "\n\n".join(doc_texts_to_send)
+                state["messages"].append({"role": "user", "content": doc_question})
+                state["streaming"] = True
+                state["stream_buffer"] = ""
+                save_session_state(session_id)
+                def run_stream_for_doc(session_id, messages, doc_question):
+                    try:
+                        system_prompt = load_system_prompt()
+                        rag_chunks = []
+                        try:
+                            global_embeds = []
+                            global_texts = []
+                            global_fnames = []
+                            for fname, emb in EMBEDDING_INDEX.items():
+                                global_embeds.append(emb)
+                                global_texts.append(EMBEDDING_TEXTS[fname])
+                                global_fnames.append(fname)
+                            global_rag = semantic_search(doc_question, global_embeds, global_texts, global_fnames, top_k=2)
+                            if global_rag:
+                                for r in global_rag:
+                                    rag_chunks.append(f"Global doc [{r['filename']}]:\n{r['text'][:1000]}")
+                            user_embeds, user_texts, user_fnames = get_user_embeddings(session_id)
+                            user_rag = semantic_search(doc_question, user_embeds, user_texts, user_fnames, top_k=2)
+                            if user_rag:
+                                for r in user_rag:
+                                    rag_chunks.append(f"User upload [{r['filename']}]:\n{r['text'][:1000]}")
+                        except Exception as e:
+                            logger.error(f"Session {session_id}: RAG error (doc upload): {e}")
+                        context_block = ""
+                        if rag_chunks:
+                            context_block = "The following sources may help answer the question:\n\n" + "\n\n".join(rag_chunks) + "\n\n"
+                        msg_list = [{"role": "system", "content": system_prompt}]
+                        if context_block:
+                            msg_list.append({"role": "system", "content": context_block})
+                        for m in messages:
+                            msg_list.append({"role": m["role"], "content": m["content"]})
+                        response = openai.ChatCompletion.create(
+                            model="gpt-3.5-turbo",
+                            messages=msg_list,
+                            max_tokens=700,
+                            temperature=0.2,
+                            stream=True,
+                        )
+                        reply = ""
+                        for chunk in response:
+                            delta = chunk["choices"][0]["delta"]
+                            content = delta.get("content", "")
+                            if content:
+                                reply += content
+                                session_lock = get_session_lock(session_id)
+                                with session_lock:
+                                    load_session_state(session_id)
+                                    state = get_session_state(session_id)
+                                    state["stream_buffer"] = reply
+                                    save_session_state(session_id)
+                        session_lock = get_session_lock(session_id)
+                        with session_lock:
+                            load_session_state(session_id)
+                            state = get_session_state(session_id)
+                            state["messages"].append({"role": "assistant", "content": reply})
+                            state["stream_buffer"] = ""
+                            state["streaming"] = False
+                            save_session_state(session_id)
+                        logger.info(f"Session {session_id}: Assistant responded to doc upload")
+                    except Exception as e:
+                        session_lock = get_session_lock(session_id)
+                        with session_lock:
+                            load_session_state(session_id)
+                            state = get_session_state(session_id)
+                            state["streaming"] = False
+                            state["stream_buffer"] = ""
+                            save_session_state(session_id)
+                        logger.error(f"Session {session_id}: Streaming error for doc upload: {e}")
+                threading.Thread(target=run_stream_for_doc, args=(session_id, list(state["messages"]), doc_question), daemon=True).start()
+                start_streaming = True
 
         send_triggered = False
         if trigger == "send-btn" or trigger == "hidden-send":