Dataherald is an open-source engine for interacting with SQL databases in natural language: ask a question in plain English and it generates, runs and explains the SQL for you. It was built as a modular natural-language-to-SQL platform that companies can embed into their own products, with fine-tuning support to adapt the model to a specific database schema and business vocabulary. The project has around 3,650 stars on GitHub and is Apache-2.0 licensed.
Dataherald’s architecture separates the NL-to-SQL pipeline into swappable components: a database connector, a vector store holding schema context and example queries, and an LLM agent that plans and writes the query. It connects to warehouses and databases such as PostgreSQL, Snowflake, BigQuery and Databricks, and returns both the generated SQL and the results so users can verify what ran. Administrators can add “golden” example queries and business context to steadily improve accuracy, and the engine exposes a REST API so the capability can be embedded behind a chat interface or dashboard. A fine-tuning path lets teams train smaller models on their own query history.
The Dataherald engine is free and open source under the Apache-2.0 license for self-hosting. The company behind it offered hosted and enterprise deployments, so it is listed as Freemium; self-hosters only pay for their own LLM API usage.
Data teams and SaaS builders who want to add natural-language querying to a product, and analysts who would rather ask questions in English than hand-write SQL against large warehouse schemas.