What is Tabletalk

tabletalk is a declarative language and command-line interface (CLI) tool designed for seamless interaction with databases, allowing users to define data access configurations in a YAML file and query data using natural language.

Use cases

Use cases for tabletalk include simplifying database queries for non-technical users, enabling rapid prototyping of data-driven applications, and enhancing data exploration processes by allowing users to ask questions in plain English.

How to use

To use tabletalk, install it via pip, configure the tabletalk.yaml file with your database and LLM preferences, and define contexts for your data. You can then run queries in natural language, and tabletalk will generate the corresponding SQL.

Key features

Key features of tabletalk include support for multiple databases (BigQuery, SQLite, MySQL, Postgres), the ability to define custom contexts for focused querying, integration with large language models (LLMs) for SQL generation and answering questions, and the capability to execute generated SQL locally.

Where to use

tabletalk can be used in various fields such as data analysis, business intelligence, and application development where database interaction is required, especially for users who prefer natural language queries over traditional SQL.

tabletalk

tabletalk is a command-line interface (CLI) tool designed to let you “talk” to your databases using natural language. Unlike heavier frameworks, tabletalk is built for simplicity and ease of use. With tabletalk, you can define specific “contexts” based on relationships in your data, then query that data conversationally, either by generating SQL or asking questions directly. It connects to your existing databases—BigQuery, SQLite, MySQL, or Postgres—pulls schemas based on your defined contexts, and leverages large language models (LLMs) from OpenAI and Anthropic to chat with your data effectively.

Features

• Database Support: Connect to BigQuery, SQLite, MySQL, and Postgres.
• Custom Contexts: Define relationships in your data to create focused querying scenarios.
• LLM Integration: Use OpenAI or Anthropic models to generate SQL or answer questions.
• Natural Language Queries: Ask questions about your data in plain English, with SQL generated automatically.
• Local Execution: Run generated SQL locally against your database.

Installation

Install tabletalk via pip:

pip install tabletalk

Configuration

tabletalk relies on a configuration file named tabletalk.yaml to set up your database and LLM preferences. This file includes:

• Provider: Details for connecting to your database.
• LLM: Settings for the language model, such as provider, API key, and model specifics.
• Contexts: Path to a directory containing context definitions.
• Output: Directory where manifest files (schema data) are stored.

Note: For security, set API keys as environment variables (e.g., export ANTHROPIC_API_KEY="your-key-here").

Example tabletalk.yaml:

provider:
  type: mysql
  host: localhost
  user: root
  password: ${MYSQL_PASSWORD}
  database: test_store

llm:
  provider: anthropic
  api_key: ${ANTHROPIC_API_KEY}
  model: claude-3-5-sonnet-20240620
  max_tokens: 500
  temperature: 0

contexts: contexts
output: manifest

Defining Contexts

Contexts are defined in separate YAML files within the contexts/ directory. Each context specifies a subset of your database—datasets and tables—relevant to a particular querying scenario.

Example contexts/sales_context.yaml:

name: sales_context
datasets:
  - name: test_store
    tables:
      - customers
      - orders

Usage

tabletalk offers three core CLI commands:

Initialize the Project

tabletalk init

Creates tabletalk.yaml, a contexts/ folder, and a manifest/ folder.

Apply Contexts

tabletalk apply

Reads context definitions, connects to your database, pulls the relevant schemas, and generates manifest files in the manifest/ directory.

Query Command

Starts an interactive session for querying your data via the command line.

tabletalk query [PROJECT_FOLDER]

• PROJECT_FOLDER: (Optional) Path to the project directory. Defaults to the current directory.

Steps:

1. Select a Manifest: Choose from available manifest files (e.g., 1. sales_context.json).
2. Ask Questions: Type a natural language question (e.g., “How many customers placed orders last month?”).
3. Change Manifests: Type change to switch to a different manifest.
4. Exit: Type exit to end the session.

Serve Command

Launches a Flask web server providing a graphical interface for querying your data.

tabletalk serve [--port PORT]

• –port PORT: (Optional) Specifies the port. Defaults to 5000.

Steps:

1. Open the Web Interface: Navigate to http://localhost:PORT.
2. Select a Manifest: Click a manifest (e.g., sales_context.json).
3. Ask a Question: Type a question (e.g., “How many customers placed orders last month?”) and click “Send”.

Note: Both query and serve commands require manifest files, generated by running tabletalk apply first.

Example Workflow

Step 1: Initialize the Project

tabletalk init

Creates the following structure:

project_folder/
├── tabletalk.yaml
├── contexts/
└── manifest/

Step 2: Define a Context

Create contexts/sales_context.yaml:

name: sales_context
datasets:
  - name: test_store
    tables:
      - customers
      - orders

Step 3: Apply the Schema

tabletalk apply

Generates a manifest file (e.g., manifest/sales_context.json).

Step 4: Query Your Data

tabletalk query

1. Select sales_context.json.
2. Ask a question like: “How many customers placed orders last month?”
3. View the generated SQL and execute it locally.
4. Type exit to end the session.

Step 5: Start the Web Server

tabletalk serve --port 8080

Step 6: Access the Web Interface

Open http://localhost:8080 in your browser.

• Select a Manifest: Click on sales_context.json.
• Ask a Question: Type “How many customers placed orders last month?” and click “Send”.
• View Generated SQL: The query appears in the chat history.

Contributing

Want to help improve tabletalk? Fork the repository, make your changes, and submit a pull request. For major updates, please open an issue first to discuss your ideas.

License

This code is licensed under CC BY-NC 4.0 for non-commercial use. For commercial use, contact [email protected].