updated output

fiscal.py

@@ -76,7 +76,7 @@ QA_TEMPLATE = """You are FISCAL, a friendly AI assistant for Canadian banking, p
 The user has connected their bank account. Here is their live financial data:
 {financial_context}
 
-Answer using ONLY the context below combined with the financial data above. Do not add facts not in the context. Do not mention any underlying AI model. If the context does not contain enough info, say so briefly and offer to help with a related Canadian finance topic. If the question is a greeting or casual message, respond warmly and briefly introduce yourself as FISCAL, a Canadian personal finance assistant. Otherwise, answer using the context below combined with the financial data above. Do not add facts not in the context. Do not mention any underlying AI model. If the context does not contain enough information, say so briefly and offer to help with a related Canadian finance topic.
+Answer using ONLY the context below combined with the financial data above. Do not add facts not in the context. Do not mention any underlying AI model. If the context does not contain enough info, say so briefly and offer to help with a related Canadian finance topic. If the question is a greeting or casual message, respond warmly and briefly introduce yourself as FISCAL, a Canadian personal finance assistant. Otherwise, answer using the context below combined with the financial data above. Do not add facts not in the context. Do not mention any underlying AI model. Do not mention Plaid, access tokens, sandbox, or any technical errors. If the context does not contain enough information, say so briefly and offer to help with a related Canadian finance topic. Do not add facts not in the context. Do not mention any underlying AI model. If the context does not contain enough information, say so briefly and offer to help with a related Canadian finance topic.
 
 Context:
 {context}
@@ -100,10 +100,16 @@ def _get_memory(user_id: str) -> ConversationBufferMemory:
 
 
 def get_answer(message: str, user_id: str, financial_context: str = "") -> str:
-    # Build prompt with financial context already filled in
+
+    # Clean financial context — never expose raw errors to the LLM
+    if not financial_context or financial_context == "NO_BANK_DATA" or "Unable to fetch" in financial_context or "Bad Request" in financial_context or "access token" in financial_context.lower():
+        context_block = "The user has not connected their bank account yet. If relevant, gently mention they can connect their bank account in the Options menu to get personalized advice based on their real transactions and balances."
+    else:
+        context_block = f"The user has connected their bank account. Here is their live financial data:\n{financial_context}"
+
     filled_template = QA_TEMPLATE.replace(
         "{financial_context}",
-        financial_context or "No bank data available."
+        context_block
     )
 
     prompt = PromptTemplate(
@@ -123,7 +129,6 @@ def get_answer(message: str, user_id: str, financial_context: str = "") -> str:
     result = chain.invoke({"question": message})
     return result["answer"]
 
-
 def clear_memory(user_id: str) -> None:
     if user_id in _memory_store:
         _memory_store[user_id].clear()

plaid_client.py

@@ -109,16 +109,12 @@ def get_transactions(access_token: str, days: int = 30) -> str:
 # ---------- Combined snapshot ----------
 
 def get_financial_snapshot(access_token: str) -> str:
-    """
-    Returns full financial context string to inject into the LLM prompt.
-    Combines balances + transaction summary.
-    """
     try:
         balances     = get_balances(access_token)
         transactions = get_transactions(access_token)
         return f"{balances}\n\n{transactions}"
     except Exception as e:
-        return f"Unable to fetch account data: {str(e)}"
+        return "NO_BANK_DATA"