More answers, less waiting
Let business teams ask questions of data in natural language and get results - when they need it.
Privacy-centric integration with any SQL database or data warehouse
Answer questions that can require pages of SQL
Automatically visualise data as tables and charts
Fine-tuned on your metadata to give results you can trust
Focus on what you do best
Handle complex and repetitive workflows in SQL, Python and R with human-in-the-loop agents.
Automates trial and error in statistical analyses
Decomposes complex, multi-part questions into executable tasks
Orchestrates handing off each task to an appropriate AI model
Generates reports for editing and collaboration
Tailored to your business
Go beyond generic models. Defog uses AI that is customized for your database schema and aligned to your business logic. Deploy fine-tuned models, monitor their performance and stay aligned with human feedback.
Keep your data private
Defog only uses metadata to fine-tune LLMs for you – it's never given access to your actual database. You can use our cloud offering or deploy models on-prem.
Powered by SQLCoder
Defog is powered by SQLCoder – the best performing open-source model for text-to-SQL generation.
State-of-the-art model
SQLCoder-34B outperforms OpenAI’s gpt-4 and gpt-4-turbo on text to SQL generation, and significantly outperforms all major open-source models for out-of-training set SQL schemas in Postgres. When fine-tuned on individual database schemas, SQLCoder-34B has 99+% accuracy for text to SQL conversions.
Our evaluation framework
SQL-Eval is our open source framework to evaluate the correctness of LLM-generated SQL. Our goal is to produce a reproducible framework that measures both the complexity and semantic diversity of queries.
Supported databases
SQLCoder is compatible with all major SQL databases and data warehouses.