Passing custom object to SQLDatabaseChain from AgentExecutor, along with input

[danishzmalik]

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

#agent executor
response = agent_executor.invoke(
            {
                "input": question, 
                "cv": str(cv)
                }, 
            config= {
                "configurable": {"session_id": f"{session_id}"}
                },
            ) 


#sql tool
sql_prompt = PromptTemplate(
    input_variables=["input", "cv", "table_info", "dialect"], template=prompt_template
)
sql_db_chain = SQLDatabaseChain.from_llm(get_llm(), db, verbose=True, top_k = 100, prompt=sql_prompt, return_intermediate_steps=True )

Description

I have created a langchain agent with mulitple tools.

These tools include a SQL tool which uses SQLDatabaseChain

Along with the input i want to pass an object called 'cv' to the agent executor, which should then pass it to the Sql tool to use in its prompt. Cv is a dict object converted to string.

My agent executor is called 'agent_executor'

response = agent_executor.invoke(
            {
                "input": question, 
                "cv": str(cv)
                }, 
            config= {
                "configurable": {"session_id": f"{session_id}"}
                },
            ) 

The prompt template of my SQL tool should look like this:

sql_prompt_template = """
You have access to the following sql tables:
{table_info}

• 'gene_expression_1' for gene expression when species is homo sapiens .
• 'gene_expression_2' for gene expression when species is mus musculus .
• 'gene_expression_3' for gene expression only when species is homo sapiens and there is a disease involved.

Given an input question:
{input}
Create a syntactically correct {dialect} query to run, then look at the results of the query and return the answer.

Use values in {cv} to generate SQl queries

Use the following format:

Question: original user question
SQLQuery: SQL Query to run
SQLResult: sql result
Answer: final answer

Question: {input}"""

I have tried the following, but it does not recognize the cv object:

sql_prompt = PromptTemplate(
    input_variables=["input", "cv", "table_info", "dialect"], template=prompt_template
)

Is there any way to achieve this without changing langchain source code?

System Info

Langchain 0.2.3

[dosubot[bot]]

To pass a custom object called cv to the SQLDatabaseChain from AgentExecutor in LangChain without changing the LangChain source code, you can use the agent_executor_kwargs parameter in the create_sql_agent function. This parameter allows you to pass arbitrary additional arguments to the AgentExecutor.

Here is an example of how you can do this:

from langchain_openai import ChatOpenAI
from langchain_community.agent_toolkits import create_sql_agent
from langchain_community.utilities import SQLDatabase

# Create your custom object
cv = CustomObject()

# Initialize the database and language model
db = SQLDatabase.from_uri("sqlite:///Chinook.db")
llm = ChatOpenAI(model="gpt-3.5-turbo", temperature=0)

# Create the SQL agent and pass the custom object via agent_executor_kwargs
agent_executor = create_sql_agent(
    llm,
    db=db,
    agent_type="tool-calling",
    verbose=True,
    agent_executor_kwargs={"cv": cv}
)

In this example, CustomObject is your custom object class. The cv object is passed to the AgentExecutor through the agent_executor_kwargs dictionary. This way, you can pass the custom object without modifying the LangChain source code ^[1].

Additionally, ensure that your PromptTemplate includes cv in the input_variables list and that the SQLDatabaseChain is correctly set up to use this prompt:

# Define the prompt template
prompt_template = """
You have access to the following sql tables:
{table_info}

- 'gene_expression_1' for gene expression when species is homo sapiens.
- 'gene_expression_2' for gene expression when species is mus musculus.
- 'gene_expression_3' for gene expression only when species is homo sapiens and there is a disease involved.

Given an input question:
{input}
Create a syntactically correct {dialect} query to run, then look at the results of the query and return the answer.

Use values in {cv} to generate SQL queries

Use the following format:

Question: original user question
SQLQuery: SQL Query to run
SQLResult: sql result
Answer: final answer

Question: {input}
"""

# Create the prompt template
sql_prompt = PromptTemplate(
    input_variables=["input", "cv", "table_info", "dialect"],
    template=prompt_template
)

# Create the SQLDatabaseChain
sql_db_chain = SQLDatabaseChain.from_llm(
    llm=get_llm(),
    db=db,
    verbose=True,
    top_k=100,
    prompt=sql_prompt,
    return_intermediate_steps=True
)

This setup ensures that the cv object is passed correctly and used within the SQL tool's prompt template.

To continue talking to Dosu, mention @dosu.

[danishzmalik]

@dosu
I have other tools as well apart from the SQL tool. The agent executor im using has a runnable sequence agent. . Not all tools are SQL tools, so i cannot use a create_sql_agent for all tools as some tools are simple QA rag tools.

[dosubot[bot]]

To pass a custom object like cv to a RunnableSequenceAgent that includes multiple tools, you can use the agent_executor_kwargs parameter to pass the custom object to the AgentExecutor. Here's how you can achieve this:

1. Define the custom object and tools:
   Create your custom object and define the tools, including the SQL tool and other tools.

2. Create the RunnableSequenceAgent:
   Use the RunnableSequenceAgent to manage the sequence of tools.

3. Pass the custom object via agent_executor_kwargs:
   Pass the custom object to the AgentExecutor using the agent_executor_kwargs parameter.

Here's an example:

from langchain_openai import ChatOpenAI
from langchain_community.utilities import SQLDatabase
from langchain_core.prompts import PromptTemplate
from langchain_core.runnables import RunnableSequenceAgent, RunnableLambda
from langchain_core.agents import AgentExecutor

# Define your custom object
cv = {"key": "value"}

# Define the SQL tool
prompt_template = """
You have access to the following sql tables:
{table_info}

- 'gene_expression_1' for gene expression when species is homo sapiens.
- 'gene_expression_2' for gene expression when species is mus musculus.
- 'gene_expression_3' for gene expression only when species is homo sapiens and there is a disease involved.

Given an input question:
{input}
Create a syntactically correct {dialect} query to run, then look at the results of the query and return the answer.

Use values in {cv} to generate SQL queries

Use the following format:

Question: original user question
SQLQuery: SQL Query to run
SQLResult: sql result
Answer: final answer

Question: {input}
"""

sql_prompt = PromptTemplate(
    input_variables=["input", "cv", "table_info", "dialect"],
    template=prompt_template
)

db = SQLDatabase.from_uri("sqlite:///Chinook.db")
llm = ChatOpenAI(model="gpt-3.5-turbo", temperature=0)

sql_db_chain = SQLDatabaseChain.from_llm(
    llm=llm,
    db=db,
    verbose=True,
    top_k=100,
    prompt=sql_prompt,
    return_intermediate_steps=True
)

# Define other tools (e.g., a simple QA tool)
def qa_tool(input):
    return f"Answer to {input}"

qa_runnable = RunnableLambda(qa_tool).as_tool(
    name="QA Tool",
    description="A simple QA tool."
)

# Create the RunnableSequenceAgent
runnable_sequence_agent = RunnableSequenceAgent(
    tools=[sql_db_chain, qa_runnable]
)

# Create the AgentExecutor and pass the custom object via agent_executor_kwargs
agent_executor = AgentExecutor(
    agent=runnable_sequence_agent,
    tools=[sql_db_chain, qa_runnable],
    verbose=True,
    agent_executor_kwargs={"cv": cv}
)

# Invoke the agent executor
response = agent_executor.invoke(
    {
        "input": "What is the gene expression for homo sapiens?",
        "cv": str(cv)
    },
    config={
        "configurable": {"session_id": "session_123"}
    }
)

print(response)

In this example:

• The cv object is passed to the AgentExecutor via agent_executor_kwargs.
• The RunnableSequenceAgent manages the sequence of tools, including the SQL tool and a simple QA tool.
• The PromptTemplate for the SQL tool includes cv in the input_variables list.

This setup ensures that the cv object is passed correctly and used within the SQL tool's prompt template, while also allowing other tools to function as needed ^[1][2][3].

To continue talking to Dosu, mention @dosu.