chatbot with RAG and Tool(function) calling

[aditya11ad]

How can we combine these two concepts:
1 chatbot with Retrieval (Ref 1)
2 chatbot with tool usage (Ref 2)

My objective:
to create a chatbot-agent capable of: Retrival from context and if function(tool) call needed then only call the tool. (while maintaining the history)

[sandhya-500]

For me also having the same objective can anybody help on this.

[abhiru-wije]

Did you resolve this issue? I have the same issue, still couldn't find the proper solution for this.

[aditya11ad]

yes,
using create_retriever_tool:
Sample Code:

#--------- 1. Retrival: ---------
retrive_tool = create_retriever_tool(
    vectorstore.as_retriever(),
    name="sports_general_chat",
    description="""Useful when you need to answer questions about sports related general queries from available FAQs.
    Not useful for answering questions about his account information.""",
) 
#--------- 2: custom tool usage: ---------
@tool
def show_account_informaion(customer_name: str) -> str:
    '''useful when customer wants to know his account related profile information.
    if he doesn't provide his own name, then ask it first. do not pass any default name in argument.
    Args:
        customer_name: customer's name associated with the account
    '''
     return "ur account have 10 credits"
    
tools = [retrive_tool, show_account_informaion]

prompt = ChatPromptTemplate.from_messages(#ur prompt.......

agent = create_openai_tools_agent(chat, tools, prompt)
agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=True)

chat_history_for_chain = ChatMessageHistory()

conversational_agent_executor = RunnableWithMessageHistory(
    agent_executor,
    lambda session_id: chat_history_for_chain,
    input_messages_key="input",
    output_messages_key="output",
    history_messages_key="chat_history",
)

Hope this will solve your usecase.
Thanks :)

[abhiru-wije]

Thanks for the answer @aditya11ad, This is helpful for my development. I will try this and let you know

[abhiru-wije]

Hi @aditya11ad,

I got some errors during the execution, I will add my code.

vectordb = Chroma.from_documents(
documents=chunked_documents,
embedding=embeddings,
collection_name=chroma_collection,
client=chroma_client,
)
vectordb.as_retriever()
system_prompt = """
Given a chat history and the latest user question
You are Agent .............

    If the user asks for account information, respond with the function `show_account_informaion`.
    """

contexualize_q_prompt = ChatPromptTemplate.from_messages(
[
("system", system_prompt),
MessagesPlaceholder("chat_history"),
("human", "{input}"),
MessagesPlaceholder(variable_name="agent_scratchpad")
]
)
llm = ChatOpenAI(model_name="gpt-4o",
api_key=OPENAI_API_KEY, temperature=0)
history_aware_retriever = create_history_aware_retriever(
llm, retrieve_tool, contexualize_q_prompt
)
qa_system_prompt = """hiiiii............"""
qa_prompt = ChatPromptTemplate.from_messages(
[
("system", qa_system_prompt),
MessagesPlaceholder("chat_history"),
("human", "{input}"),
MessagesPlaceholder(variable_name="agent_scratchpad")
]
)
question_answer_chain = create_stuff_documents_chain(llm, qa_prompt)
rag_chain = create_retrieval_chain(
history_aware_retriever, question_answer_chain)

store ={}

def get_session_history(user_id: str, conversation_id: str,) -> BaseChatMessageHistory:
global store
if (user_id, conversation_id) not in store:
store[(user_id, conversation_id)] = ChatMessageHistory()
return store[(user_id, conversation_id)]

conversational_rag_chain = RunnableWithMessageHistory(
agent_executor,
get_session_history,
input_messages_key="input",
history_messages_key="chat_history",
output_messages_key="answer",
history_factory_config=[
ConfigurableFieldSpec(
id="user_id",
annotation=str,
name="User ID",
description="Unique identifier for the user.",
default="",
is_shared=True,
),
ConfigurableFieldSpec(
id="conversation_id",
annotation=str,
name="Conversation ID",
description="Unique identifier for the conversation.",
default="",
is_shared=True,
)
],
)

answer = conversational_rag_chain.invoke(
{"input": "Hi again"},
config={
"configurable": {"user_id":"01", "conversation_id": "01"}
},)
response = answer['answer']
response

could you please tell me how to combine this one and that function calling part

[pramudithnk]

@aditya11ad

I got error like this,

"Input to ChatPromptTemplate is missing variables {'context'}. Expected: ['agent_scratchpad', 'chat_history', 'context', 'input'] Received: ['input', 'chat_history', 'intermediate_steps', 'agent_scratchpad']

Is there any way to resolve this issue

[aditya11ad]

use a prompt like this:

prompt = ChatPromptTemplate.from_messages(
    [SystemMessagePromptTemplate(prompt=PromptTemplate(input_variables=[], template='You are a helpful assistant. you need to call different tools and retrive_tool for RAG to answer sports related queries from the provided content only.  Never pass any default argument if not provided. Be specific about the answer.')),
     MessagesPlaceholder(variable_name='chat_history', optional=True),
     HumanMessagePromptTemplate(prompt=PromptTemplate(
         input_variables=['input'], template='{input}')),
     MessagesPlaceholder(variable_name='agent_scratchpad')])

[abhiru-wije]

@aditya11ad
Thanks for the support, This is working