Defog.ai

Defog.ai is an enterprise data analysis tool that lets business teams ask questions in plain language and get back SQL, tables, charts, and answers. Founded in 2023 by Medha Basu and Rishabh Srivastava and backed by Y Combinator and Script Capital, the company is best known for SQLCoder, its own family of open source large language models fine tuned for turning natural language into SQL. SQLCoder has been downloaded more than one hundred thousand times and, at points, outperformed larger general models on the company's own SQL evaluation benchmark.

Defog is built around a privacy first architecture. It never accesses or moves the underlying data. Instead it uses only metadata, such as table and column descriptions and example queries added through a cookbook, to fine tune models to a specific business. Well defined schemas can improve query accuracy substantially, and the models are designed to be interpretable and explainable so a user can trust and inspect the SQL behind an answer. Results appear first as tables and can be switched to charts or raw SQL, with an advanced mode for deeper work.

Teams can deploy Defog in the cloud or on premises to meet privacy requirements, and the open source SQLCoder model and the open source SQL evaluation framework can be run and inspected directly. Defog embeds inside an application as a conversational widget so a company's own users can ask data questions without leaving the product. It also offers agents that handle multi step workflows across SQL, Python, and R with a human in the loop, and a deep research capability for internal data. It connects to the major databases and warehouses including Postgres, Snowflake, Redshift, and Databricks.

Defog is SOC 2 Type II compliant, and its metadata only design gives it a strong privacy posture for regulated buyers. The system learns from user feedback and preferences and adapts its responses over time. A free plan includes application programming interface access, with paid tiers reported to start higher for teams and enterprises. Company records indicate a merger or acquisition event in April 2025, and while the product and its open source models remain active, the standalone commercial roadmap is less certain, so buyers should confirm current status directly.