Summary

DAT (Data Ask Tool) is an enterprise-grade AI framework enabling natural language querying of databases. It empowers business users to interact directly with data without SQL, ensuring accuracy through a pre-modeled semantic layer.

How It Works

DAT employs an "Askdata Agent workflow" prioritizing result quality. It uses LLMs for natural language understanding and semantic SQL generation, translating it into database-specific SQL. A key feature is its rich semantic modeling (entities, dimensions, measures) defined via YAML, which guides LLM precision. Vectorized retrieval enhances query understanding with stored knowledge.

Quick Start & Requirements

• Requirements: Java 17+, a supported database (MySQL, PostgreSQL, Oracle, DuckDB), and an LLM API key (OpenAI, Anthropic, etc.).
• Quick Start: Install DAT CLI. Initialize a project (dat init), configure dat_project.yaml (DB/LLM), define semantic models in YAML, and run queries via dat run or dat server openapi.

Highlighted Details

• Enterprise Architecture: Pluggable SPI, modular design, factory pattern.
• Multi-Database Support: Native MySQL, PostgreSQL, Oracle, DuckDB; extensible via SPI.
• Semantic SQL Generation: LLM-driven NLU, SQL dialect conversion, semantic model binding.
• Semantic Modeling: YAML-defined entities, dimensions (time, categorical), measures.
• Vectorized Retrieval: Enhances queries with embeddings for SQL Q&A, synonyms, knowledge.
• Flexible Deployment: CLI, OpenAPI service, MCP service for agent integration.

Maintenance & Community

Actively developed, with a detailed "Development Items List" outlining planned features. Community engagement via GitHub Discussions and WeChat.

Licensing & Compatibility

Licensed under Apache 2.0, permissive for commercial use and closed-source integration.

Limitations & Caveats

Key features under development include IDE plugins (VSCode, IDEA, Eclipse), LLM-assisted semantic model generation, comprehensive testing for core querying, and Jinja templating for data permissions. The project is at version 0.7.2, indicating ongoing development.