Ask RITA (Reasoning Interface for Text-to-Analytics)

Ask what. Get answers. RITA turns a natural-language question into SQL, statistics, and insights — no code required.

Go beyond simple text-to-SQL. Ask RITA is an LLM-powered analytics framework that generates queries, runs scipy-backed statistical tests, conducts CRISP-DM research workflows, classifies data, and visualizes results — across SQL and NoSQL databases — from a single natural-language question.

🔒 IMPORTANT — Read-Only Database Access Required

AskRITA generates and executes SQL/NoSQL queries against your database. LLM-generated queries are inherently unpredictable. To prevent inadvertent writes, deletes, or schema changes:

1. Always connect with a read-only database user. Grant only SELECT (SQL) or find/aggregate (MongoDB) permissions. Never use credentials with INSERT, UPDATE, DELETE, DROP, or DDL privileges.
2. Do not rely on application-level safeguards alone. AskRITA includes prompt-injection detection and blocks known destructive patterns, but these are defence-in-depth measures — not substitutes for proper database permissions.
3. Store credentials in environment variables (${DB_USER}, ${DB_PASSWORD}), never in config files. See Configuration Guide.

The database user's granted permissions are the only reliable boundary between AskRITA and your data.

🚀 Four Powerful Workflows

📊 SQLAgentWorkflow - Natural Language to SQL

• 🗣️ Natural Language to SQL - Ask questions in plain English
• 💬 Conversational Queries - Follow-up questions with context awareness
• 🗄️ Multi-Database Support - PostgreSQL, MySQL, SQLite, SQL Server, BigQuery, Snowflake, IBM Db2
• 📊 Smart Visualization - Automatic chart recommendations
• 🔄 Error Recovery - Automatic SQL retry with error feedback

🍃 NoSQLAgentWorkflow - Natural Language to MongoDB

• 🗣️ Natural Language to MongoDB - Ask questions, get aggregation pipelines
• 🍃 MongoDB Support - mongodb:// and mongodb+srv:// (Atlas) connections
• 🛡️ Safety Validation - Blocks destructive operations, read-only analytics
• 🔄 Full Feature Parity - PII detection, visualization, follow-up questions, Chain-of-Thoughts

🔬 ResearchAgent - CRISP-DM Data Science Research

• 📋 CRISP-DM Methodology - Complete 6-phase data science workflow
• 🧪 Hypothesis Testing - Automated research question formulation and testing
• 📊 Real Statistics - scipy-powered t-tests, ANOVA, correlation, chi-square (not LLM-generated!)
• 📈 Effect Sizes - Cohen's d, η², Cramér's V with automatic interpretation
• 🎯 Actionable Insights - Data-driven recommendations with confidence levels

🏷️ DataClassificationWorkflow - LLM-Powered Data Processing

• 🖼️ Image Classification - AI extracts data directly from images (medical bills, invoices, documents)
• 📄 Excel/CSV Processing - Process large datasets with AI classification
• 🚀 API-First Design - Perfect for microservices with dynamic field definitions per request
• 🧠 Multi-Tenant Support - Different schemas per customer/organization without server restarts

📊 Model Performance Comparison (BIRD Benchmark)

BIRD Mini-Dev text-to-SQL execution accuracy (EX) across 500 questions, with oracle knowledge (evidence) enabled.

Core Features

• 🤖 Multi-Cloud LLM Integration - OpenAI, Azure, Google Cloud Vertex AI, AWS Bedrock
• ⚙️ Configurable Workflows - Enable/disable steps, customize prompts, enhanced security options
• 🔒 Enterprise Security - Credential management, access controls, audit logging
• 🛡️ PII/PHI Detection - Automatic privacy protection with Microsoft Presidio analyzer
• 🏗️ Production Ready - Design pattern architecture, comprehensive logging, error handling, monitoring
• 🌐 Advanced BigQuery - Cross-project dataset access, 3-step validation, configurable access patterns
• 📊 Token Management - Built-in token utilities for cost optimization and LLM efficiency
• 🧪 Extensive Testing - Full test suite with quality assurance tools (550+ tests passing)
• 🔌 Type-Safe Integration - Exported Pydantic models for seamless downstream application integration

Quick Start

1. Install

pip install askrita

📋 More options: Installation Guide — pip, Poetry, from-source, development setup

2. Configure

export OPENAI_API_KEY="your-api-key-here"
cp example-configs/query-openai.yaml my-config.yaml

⚙️ Full reference: Configuration Guide

3. Use

from askrita import SQLAgentWorkflow, ConfigManager

config = ConfigManager("my-config.yaml")
workflow = SQLAgentWorkflow(config)
result = workflow.query("What are the top 10 customers by revenue?")
print(result['answer'])

NoSQL (MongoDB)

from askrita import NoSQLAgentWorkflow, ConfigManager

config = ConfigManager("mongodb-config.yaml")
workflow = NoSQLAgentWorkflow(config)
result = workflow.query("How many orders were placed last month?")
print(result.answer)

Research Agent - CRISP-DM Data Science

from askrita import ConfigManager
from askrita.research import ResearchAgent

config = ConfigManager("my-config.yaml")
research = ResearchAgent(config)

result = research.test_hypothesis(
    research_question="How does customer satisfaction differ across business lines?",
    hypothesis="Medicare members have higher NPS scores than Commercial members"
)

print(f"Conclusion: {result['conclusion']}")  # SUPPORTED, REFUTED, or INCONCLUSIVE
print(f"P-value: {result['key_metrics'].get('p_value')}")  # Real scipy computation

📖 All examples: Usage Examples & API Reference — conversational queries, data classification, exports, CLI, result format

⚠️ Important: Configuration file with LLM provider settings and prompts is always required. API keys are read from environment variables.

Type-Safe Integration

from askrita import (
    SQLAgentWorkflow, ConfigManager,
    UniversalChartData, ChartDataset, DataPoint, WorkflowState
)

result: WorkflowState = workflow.query("Show me sales by region")
chart = UniversalChartData(**result['chart_data'])

Supported Platforms

Databases: PostgreSQL, MySQL, SQLite, SQL Server, BigQuery, Snowflake, IBM DB2, MongoDB

LLM Providers: OpenAI, Azure OpenAI, Google Cloud Vertex AI, AWS Bedrock

📋 Connection strings, auth details, config templates: Supported Platforms

Configuration

Required Components

Quick Setup

export OPENAI_API_KEY="your-api-key-here"
cp example-configs/query-openai.yaml my-config.yaml

Configuration Templates

example-configs/query-openai.yaml           # OpenAI + PostgreSQL
example-configs/query-azure-openai.yaml     # Azure OpenAI
example-configs/query-snowflake.yaml        # Snowflake database
example-configs/query-mongodb.yaml          # MongoDB (NoSQL)
example-configs/example-zscaler-config.yaml # Corporate proxy setup
example-configs/data-classification-*.yaml  # Data processing workflows

📚 Complete reference: Configuration Guide

Corporate Proxy & SSL

llm:
  ca_bundle_path: "./credentials/zscaler-ca.pem"

📚 Full guide: CA Bundle Setup

MCP Server (for AI Assistants)

{
  "mcpServers": {
    "askrita": {
      "command": "askrita",
      "args": ["mcp", "--config", "/path/to/your/config.yaml"]
    }
  }
}

📖 Setup guide: Claude Desktop Setup

Development

Setup

git clone https://github.com/cvs-health/askRITA.git
cd askRITA
pip install poetry && poetry install

Quality Checks

poetry run pytest                    # Tests
poetry run black askrita/         # Format  
poetry run flake8 askrita/        # Lint
poetry run mypy askrita/          # Type check

📚 Documentation

📖 Complete index: Documentation Site

License

Apache License 2.0 - see LICENSE file for details.