Natural query

Natural query is a tool that leverages you favorite large language models (LLMs) to translate natural language queries into SQL. It further augments the answers using the same LLM. This tool is highly beneficial for those who need to interact with databases in natural language.

Get started

Install the library:

git clone https://github.com/deltawi/NaturalQuery.git
cd NaturalQuery
pip install -e .

Example

from naturalquery.query_translator import QueryTranslator

# Initialize the translator with the configuration file
query_translator = QueryTranslator(config_path="config.yaml")

# Translate and get an answer to a natural language query
query_translator.answer("Who are the suppliers with products that have the most reviews from costumers ?")

The supplier with products that has the most reviews from customers is "Global Supplies" with a review count of 1.

Features

• Natural Language to SQL Conversion: Translates questions posed in natural language to SQL queries.
• Answer Augmentation: Enhances the responses using the same LLM.
• Support for Multiple LLM Providers: Compatible with Anyscale, Cohere, OpenAI, and Custom API endpoints.
• Configurable: Allows customization through a config.yaml file.

Configuration

LLM Providers

• anyscale, cohere, openai, or custom for custom endpoints.
• API keys and model specifications are defined in the config.yaml file.

llm:
  provider: anyscale
  api_key: <your_api_key>
  model: mistralai/Mixtral-8x7B-Instruct-v0.1
  model_kwargs:
    temperature: 0.7
    #max_tokens: 8000
database:
  provider: postgres
  dbname: postgres
  host: localhost
  port: 5432
  user: postgres
  password: <your_password>
  # user_env: USR_ENV_VAR
  # password_env: PWD_ENV_VAR

Database Configuration

• Currently supports PostgreSQL, SQLite and SQLServer.
• Database connection details (host, port, user, password) are specified in the configuration file.

SQLite Config

database:
  provider: sqlite
  database: path/to/your/sqlite.db

SQL Server Config

database:
  provider: sqlserver
  driver: <sql_server_driver>
  server: <server_name>
  database: <database_name>
  username: <username>
  password: <password>

Custom Endpoint Configuration

• For custom LLM providers, specify the url and additional model_kwargs as needed.

llm:
  provider: custom
  api_key: somekey
  url: http://localhost:1234/v1/chat/completions
  model_kwargs:
    temperature: 0.7
    stream: False

Usage

1. Import the QueryTranslator class from the package.
2. Initialize QueryTranslator with the path to your configuration file.
3. Call the answer method with your natural language query.
4. The response will be printed, which includes the SQL translation and augmented answer.

Example

from query_translator.translator import QueryTranslator

# Initialize the translator with the configuration file
query_translator = QueryTranslator(config_path="config.yaml")

# Translate and get an answer to a natural language query
answer = query_translator.answer("Who are the suppliers with products that have the most reviews from costumers ?")
print(answer)

The supplier with products that has the most reviews from customers is "Global Supplies" with a review count of 1.

Using French language

from query_translator.translator import QueryTranslator

# Initialize the translator with the configuration file
query_translator = QueryTranslator(config_path="config.yaml", language="Fr")
query_translator.answer("Quels sont les fournisseurs qui ont des produits les plus notés par les clients ?")

 Le fournisseur avec le plus de produits évalués par les clients est Global Supplies (supplierid = 1, avec 1 évaluation). Il est suivi de Quality Goods Inc. (supplierid = 2, avec également 1 évaluation).