new prompt

agent.py

@@ -18,67 +18,59 @@ from supabase.client import Client, create_client
 
 load_dotenv()
 
+# === Tools ===
+
 @tool
 def multiply(a: int, b: int) -> int:
-    """Multiply two integers and return the result."""
+    """Multiply two integers."""
     return a * b
 
 @tool
 def add(a: int, b: int) -> int:
-    """Add two integers and return the result."""
+    """Add two integers."""
     return a + b
 
 @tool
 def subtract(a: int, b: int) -> int:
-    """Subtract the second integer from the first and return the result."""
+    """Subtract b from a."""
     return a - b
 
 @tool
 def divide(a: int, b: int) -> float:
-    """Divide the first integer by the second and return the result as float."""
+    """Divide a by b."""
     if b == 0:
         raise ValueError("Cannot divide by zero.")
     return a / b
 
 @tool
 def modulus(a: int, b: int) -> int:
-    """Return the remainder when the first integer is divided by the second."""
+    """Return a modulo b."""
     return a % b
 
 @tool
 def wiki_search(query: str) -> str:
-    """Search Wikipedia for a query and return up to 2 results."""
+    """Search Wikipedia for a query."""
     search_docs = WikipediaLoader(query=query, load_max_docs=2).load()
-    formatted = "\n\n---\n\n".join(
-        [f'<Document source="{doc.metadata["source"]}" page="{doc.metadata.get("page", "")}"/>\n{doc.page_content}\n</Document>' for doc in search_docs]
-    )
-    return {"wiki_results": formatted}
+    return "\n\n---\n\n".join([doc.page_content for doc in search_docs])
 
 @tool
 def web_search(query: str) -> str:
-    """Search Tavily for a query and return up to 3 results."""
+    """Search the web for a query."""
     search_docs = TavilySearchResults(max_results=3).invoke(query=query)
-    formatted = "\n\n---\n\n".join(
-        [f'<Document source="{doc.metadata["source"]}" page="{doc.metadata.get("page", "")}"/>\n{doc.page_content}\n</Document>' for doc in search_docs]
-    )
-    return {"web_results": formatted}
+    return "\n\n---\n\n".join([doc.page_content for doc in search_docs])
 
 @tool
 def arvix_search(query: str) -> str:
-    """Search Arxiv for a query and return up to 3 results."""
+    """Search Arxiv for a query."""
     search_docs = ArxivLoader(query=query, load_max_docs=3).load()
-    formatted = "\n\n---\n\n".join(
-        [f'<Document source="{doc.metadata["source"]}" page="{doc.metadata.get("page", "")}"/>\n{doc.page_content[:1000]}\n</Document>' for doc in search_docs]
-    )
-    return {"arvix_results": formatted}
+    return "\n\n---\n\n".join([doc.page_content[:1000] for doc in search_docs])
 
-# Load system prompt
+# === System Prompt ===
 with open("system_prompt.txt", "r", encoding="utf-8") as f:
     system_prompt = f.read()
-
 sys_msg = SystemMessage(content=system_prompt)
 
-# Setup Supabase vector retriever
+# === Embeddings and Vector Store ===
 embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-mpnet-base-v2")
 supabase: Client = create_client(os.environ.get("SUPABASE_URL"), os.environ.get("SUPABASE_SERVICE_KEY"))
 vector_store = SupabaseVectorStore(
@@ -88,19 +80,10 @@ vector_store = SupabaseVectorStore(
     query_name="match_documents_langchain",
 )
 
-create_retriever_tool = create_retriever_tool(
-    retriever=vector_store.as_retriever(),
-    name="Question Search",
-    description="A tool to retrieve similar questions from a vector store."
-)
-
-# Define tool list
-tools = [
-    multiply, add, subtract, divide, modulus,
-    wiki_search, web_search, arvix_search
-]
+# === Tools ===
+tools = [multiply, add, subtract, divide, modulus, wiki_search, web_search, arvix_search]
 
-# Build graph
+# === Build Graph ===
 def build_graph(provider: str = "groq"):
     if provider == "google":
         llm = ChatGoogleGenerativeAI(model="gemini-2.0-flash", temperature=0)
@@ -119,11 +102,15 @@ def build_graph(provider: str = "groq"):
     llm_with_tools = llm.bind_tools(tools)
 
     def assistant(state: MessagesState):
-        return {"messages": [llm_with_tools.invoke(state["messages"])]}
+        response = llm_with_tools.invoke(state["messages"])
+        content = response.content.strip()
+        if "FINAL ANSWER:" in content:
+            content = content.split("FINAL ANSWER:")[-1].strip()
+        return {"messages": [AIMessage(content=content)]}
 
     def retriever(state: MessagesState):
-        similar = vector_store.similarity_search(state["messages"][0].content)
-        example_msg = HumanMessage(content=f"Here I provide a similar question and answer for reference: \n\n{similar[0].page_content}")
+        similar_question = vector_store.similarity_search(state["messages"][0].content)
+        example_msg = HumanMessage(content=f"Reference: {similar_question[0].page_content}")
         return {"messages": [sys_msg] + state["messages"] + [example_msg]}
 
     builder = StateGraph(MessagesState)

system_prompt.txt

@@ -1,11 +1,6 @@
-You are a helpful assistant tasked with answering questions using a set of tools.
-
-You MUST strictly follow this rule: YOUR FINAL ANSWER MUST be the last line of your response and must follow this exact format:
-
-FINAL ANSWER: [your answer here]
-
-- Your final answer should be a **number** OR as few words as possible OR a comma-separated list.
-- Do not include explanations, markdown, or any additional text after FINAL ANSWER.
-- If the answer is a string, do not include articles (e.g., "the", "a") or abbreviations. Write digits in full words if requested.
-
-Any answer that does not follow the `FINAL ANSWER: ...` format exactly will be considered incorrect.
+You are a helpful assistant. Think step-by-step to solve the question. Then output only the final answer as your last message. The final answer must be:
+- a number (without comma separators or symbols),
+- a string (no articles or abbreviations),
+- or a comma-separated list of such elements.
+Do not include any explanation or prefix like "Answer:", "FINAL ANSWER:", or similar.
+Return only the answer.