Modify input before it reaches a specific agent

[Ayhm-Tenderd]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

None

Description

We have a workflow with multiple agents including a Text2SQL DBAgent
We want to modify the user's input prompt only for the DBAgent if DBAgent is called, maybe using a callback or any other method.

System Info

fastapi==0.110.2
pydantic==2.0.2
pydantic-settings==2.0.3
langchain==0.2.0
langchain-core==0.2.0
langchain-openai==0.1.7
langchain-community==0.2.0
faiss-cpu==1.8.0
python-dotenv==1.0.1
uvicorn==0.23.1
psycopg2-binary==2.9.9
docx2txt==0.8
pandas==2.2.2
langgraph==0.0.55
openai==1.33.0
slack-sdk==3.30.0
SQLAlchemy==2.0.29

[dosubot[bot]]

To modify the user's input prompt specifically for the Text2SQL DBAgent in your workflow, you can use the PromptTemplate to set custom instructions for the agent. Here is an example of how you can achieve this:

from langchain_core.prompts import PromptTemplate
from langchain_openai import OpenAI
from langchain_community.utilities import SQLDatabase
from langchain.chains.sql_database.query import create_sql_query_chain

# Initialize the database and LLM as shown in the setup
db = SQLDatabase.from_uri("sqlite:///Chinook.db")
llm = OpenAI(temperature=0, verbose=True)

# Define the custom prompt template
_DEFAULT_TEMPLATE = """Given an input question, first create a syntactically correct {dialect} query to run, then look at the results of the query and return the answer.
Use the following format:

Question: "Question here"
SQLQuery: "SQL Query to run"
SQLResult: "Result of the SQLQuery"
Answer: "Final answer here"

Only use the following tables:

{table_info}

If someone asks for the table foobar, they really mean the employee table.
Question: {input}"""
PROMPT = PromptTemplate.from_template(_DEFAULT_TEMPLATE)

# Initialize the SQLDatabaseChain with the custom prompt
db_chain = create_sql_query_chain(llm, db, prompt=PROMPT, k=5)

# Example usage
result = db_chain.invoke({"question": "How many employees are there in the foobar table?"})
print(result)

This code sets up a custom prompt for the Text2SQL DBAgent, ensuring that the agent follows the specified instructions when processing user input. The _DEFAULT_TEMPLATE allows for dynamic insertion of the SQL dialect and the table information, making it more flexible and adaptable to different scenarios ^{[1][2][3][4][5]}.

Additionally, LangChain provides examples of modifying input prompts for specific agents using callbacks. For instance, the create_react_agent method allows for modifying the graph state before the language model is called using the state_modifier parameter. Here is an example:

from langchain_core.messages import SystemMessage
from langgraph.prebuilt import create_react_agent

system_message = "You are a helpful assistant. Respond only in Spanish."
app = create_react_agent(model, tools, state_modifier=system_message)

messages = app.invoke({"messages": [("user", query)]})

You can also pass an arbitrary function to modify the state messages:

from langgraph.prebuilt import create_react_agent
from langgraph.prebuilt.chat_agent_executor import AgentState
from langchain_core.prompts import ChatPromptTemplate

prompt = ChatPromptTemplate.from_messages(
    [
        ("system", "You are a helpful assistant. Respond only in Spanish."),
        ("placeholder", "{messages}"),
    ]
)

def _modify_state_messages(state: AgentState):
    return prompt.invoke({"messages": state["messages"]}).to_messages() + [
        ("user", "Also say 'Pandamonium!' after the answer.")
    ]

app = create_react_agent(model, tools, state_modifier=_modify_state_messages)

messages = app.invoke({"messages": [("human", query)]})
print(
    {
        "input": query,
        "output": messages["messages"][-1].content,
    }
)

These examples demonstrate how you can modify the input prompt dynamically for specific agents using callbacks ^[6].

To continue talking to Dosu, mention @dosu.