LangChain + OpenAI + Azure SQL

.env

OPENAI_API_KEY=
OPENAI_API_BASE=
DATABASE_USERNAME=
DATABASE_PASSWORD=
DATABASE_SERVER=
DATABASE_DB=

devcontainer.json

// PUT THIS IN .devcontainer FOLDER!
// For format details, see https://aka.ms/devcontainer.json. For config options, see the
// README at: https://github.com/devcontainers/templates/tree/main/src/python
{
	"name": "Python 3",
	// Or use a Dockerfile or Docker Compose file. More info: https://containers.dev/guide/dockerfile
	"image": "mcr.microsoft.com/devcontainers/python:0-3.11",
	"features": {
		"ghcr.io/devcontainers/features/python:1": {
			"installJupyterlab": true
		},
		"ghcr.io/devcontainers/features/azure-cli:1": {}
	},

	// Features to add to the dev container. More info: https://containers.dev/features.
	// "features": {},

	// Use 'forwardPorts' to make a list of ports inside the container available locally.
	// "forwardPorts": [],

	// Use 'postCreateCommand' to run commands after the container is created.
	"postCreateCommand": "pip3 install --user -r requirements.txt",
	"customizations": {
		"vscode": {
			"extensions": [
				"ms-python.python",
				"ms-toolsai.jupyter"
			]
		}
	}

	// Configure tool-specific properties.
	// "customizations": {},

	// Uncomment to connect as root instead. More info: https://aka.ms/dev-containers-non-root.
	// "remoteUser": "root"
}

langchain_sql.py

import os
from langchain.llms import AzureOpenAI
import openai
from langchain.agents import create_sql_agent
from langchain.agents.agent_toolkits import SQLDatabaseToolkit
from langchain.sql_database import SQLDatabase
from dotenv import load_dotenv

load_dotenv()

# Configure OpenAI API
openai.api_type = "azure"
openai.api_version = "2022-12-01"
openai.api_base = os.getenv('OPENAI_API_BASE')
openai.api_key = os.getenv("OPENAI_API_KEY")

llm = AzureOpenAI(deployment_name="text-davinci-003", model_name="text-davinci-003")

database_user = os.getenv("DATABASE_USERNAME")
database_password = os.getenv("DATABASE_PASSWORD")
database_server = os.getenv("DATABASE_SERVER")
database_db = os.getenv("DATABASE_DB")

connection_string = f"mssql+pymssql://{database_user}:{database_password}@{database_server}.database.windows.net:1433/{database_db}"

db = SQLDatabase.from_uri(connection_string)
toolkit = SQLDatabaseToolkit(db=db, llm=llm, reduce_k_below_max_tokens=True)

agent_executor = create_sql_agent(
    llm=llm,
    toolkit=toolkit,
    verbose=True
)

db = SQLDatabase.from_uri("mssql+pymssql://<username>:<password>@<server>.database.windows.net:1433/<database>")
toolkit = SQLDatabaseToolkit(db=db)

agent_executor = create_sql_agent(
    llm=llm,
    toolkit=toolkit,
    verbose=True
)

agent_executor.run("Find the 5 most expensive records!")

requirements.txt

openai
langchain
pymssql
python-dotenv

Create an Azure SQL database and upload some data! Suggest this one: https://github.com/lerocha/chinook-database

seems like by default, the LLM generate SQL with mysql syntax - for example SELECT * FROM cache_instances LIMIT 10. but that does't work in MS SQL database. MS SQL should be something like select top 10 * from cache_instances. Any idea how to specific database type?

Hello Andreas, I am getting error
openai\api_requestor.py", line 683, in _interpret_response_line
raise self.handle_error_response(
openai.error.InvalidRequestError: Resource not found

Yes I am also getting the same error.

@jshang11, doesn't it self-correct after the error? https://python.langchain.com/en/latest/modules/chains/examples/sqlite.html#sql-chain-example gives more information, probably setting the query checker does the trick.

@UriDimant @rajaniesh did you check that there's a deployment with "text-davinci-003" in you Azure OpenAI instance?

Hi, I was struggling with this too, but I could resolve it, on Azure AI Studio you can create a Deployment with a name different to the model name, if you do this, the code line llm = AzureOpenAI(deployment_name="deployment name", model_name="model name") fails with the Resource not found error, if you create the Deployment with a name exactly the same as the model name, it works. A bug? An expected behavior? Don't know but, in my case, that's the way I "fixed"