- Home
- AI Database Assistants
- Dataherald
Dataherald
What it is
Dataherald is a free, Apache-2.0 open-source natural language-to-SQL engine. Through its API, users ask plain-English questions about relational or structured data. Documented uses include helping business users query a warehouse, embedding data questions and answers in SaaS products, and creating a ChatGPT plug-in for proprietary data. The project is aimed at developers building natural-language interfaces or contributing code. Its quick start describes self-hosting with Docker and Docker Compose, with MongoDB in a separate container for application data. Listed warehouse connections are PostgreSQL, BigQuery, Databricks, and Snowflake. Included SQL-generation approaches are LangChain SQL Agent, LangChain SQL Chain, LlamaIndex SQL Generator, and Dataherald SQL Agent. A context store can hold table details, business information, and verified question-and-SQL examples. An evaluator assigns confidence scores to generated SQL. Before startup, the environment needs an OpenAI API key, organization ID, and encryption key; connection data is encrypted before Mongo storage. The repository also includes an enterprise API layer, admin console, and Slackbot.
For and against
For
- Open-source Apache-2.0 license
- Plain-English questions through an API
- Supports four listed warehouses
- Context store guides SQL generation
- Connection data is encrypted before storage
Against
- Requires self-hosting with Docker
- Needs OpenAI key and organization ID
- Generated SQL may be inaccurate without context
The verdict
Dataherald provides a configurable, self-hosted route from natural-language questions to SQL, with multiple warehouse and generation options. Developers should supply business context and review generated queries before relying on results.
Who it suits
Dataherald is for developers who want to build natural-language interfaces over structured data or extend a self-hosted project. It also fits teams embedding data Q&A into products.
On the record
- Adaptability
- The documentation describes a modular architecture whose implementations can be replaced through configuration in the .env file.dataherald.readthedocs.io · read 4 Oct 2026
- API
- The engine exposes endpoints for database connections and natural-language questions, and the quick start points to a Swagger UI at /docs.dataherald.readthedocs.io · read 8 Oct 2026
- Components
- The repository describes an engine, an enterprise API layer with authentication and organizations, an admin console for configuration and observability, and a Slackbot.github.com · read 4 Oct 2026
- Context
- The context store can use business and database context, including validated question-and-SQL examples and table information, to help generate SQL.dataherald.readthedocs.io · read 4 Oct 2026
- Context and learning
- Users can add database context by scanning tables and columns, providing verified SQL examples, and describing tables and columns; the documentation also describes active learning from usage.dataherald.readthedocs.io · read 8 Oct 2026
- Credentials
- The quick start requires an OpenAI API key, an organization ID, and an encryption key before starting the engine.dataherald.readthedocs.io · read 4 Oct 2026
- Database integrations
- The quick start says the engine supports PostgreSQL, BigQuery, Databricks, and Snowflake connections.dataherald.readthedocs.io · read 8 Oct 2026
- Database support
- The quick start lists PostgreSQL, BigQuery, Databricks, and Snowflake as currently supported warehouse connections.dataherald.readthedocs.io · read 4 Oct 2026
- Deployment
- The quick start describes self-hosting the engine with Docker, with Mongo running in a separate container for application data.dataherald.readthedocs.io · read 4 Oct 2026
- Deployment components
- The repository includes an engine, an enterprise API layer with authentication and organization and user support, an admin console for configuration and observability, and a Slack bot.github.com · read 8 Oct 2026
- Evaluation
- The module overview says an evaluator assigns a confidence score to generated SQL.dataherald.readthedocs.io · read 4 Oct 2026
- Intended users
- The documentation is intended for developers building natural-language interfaces from structured data in their own projects and contributors to the engine.dataherald.readthedocs.io · read 8 Oct 2026
- License
- The GitHub repository identifies the project as licensed under Apache-2.0.github.com · read 8 Oct 2026
- Notable accuracy consideration
- The quick start says generated SQL may be inaccurate until business logic and data context are added to the context store.dataherald.readthedocs.io · read 8 Oct 2026
- Product
- Dataherald is a natural language-to-SQL engine that exposes an API for asking questions about relational data in plain English.github.com · read 4 Oct 2026
- Requirements
- The engine requires an OpenAI API key, an organization ID, and an encryption key in its environment configuration before startup.dataherald.readthedocs.io · read 8 Oct 2026
- Security
- The quick start says database connection data is encrypted before storage in MongoDB using an ENCRYPT_KEY and recommends changing the default MongoDB username and password.dataherald.readthedocs.io · read 8 Oct 2026
- Self-hosting
- The quick start describes running the engine locally with Docker and Docker Compose, alongside MongoDB for application data.dataherald.readthedocs.io · read 8 Oct 2026
