Vanna AI

Vanna AI is an open-source framework that enables natural-language-to-SQL querying across multiple databases, allowing users of all technical levels to interact with relational data through conversational interfaces. The core product, Vanna 2.0, provides a ready-to-use agent called fSQL Agent that accepts plain English questions, translates them into SQL queries, executes them against connected databases, and returns structured results or visualizations. The framework supports multi-turn dialogues, meaning users can ask follow-up questions, refine queries, or explore data iteratively without rewriting SQL. Access controls are built-in, allowing administrators to scope query execution to specific user roles, tables, or row-level permissions, ensuring that sensitive data remains protected. Vanna integrates with any large language model provider, including OpenAI, Anthropic, Cohere, and local models via Ollama or Hugging Face, and supports a wide range of database backends such as PostgreSQL, MySQL, Snowflake, BigQuery, Databricks, Redshift, SQLite, DuckDB, and others. The system can be deployed in two modes: as a cloud-hosted service managed by Vanna, or as a self-hosted solution where the entire stack runs on the user’s infrastructure. For self-hosting, the framework provides Docker images, Python SDK, and REST APIs that can be embedded into existing applications or used as standalone microservices. The agent’s UI components are rich and customizable, allowing developers to build chat-based data exploration tools, dashboards, or embedded analytics widgets. Quota management is built-in, enabling teams to track usage per user or per API key, set rate limits, and monitor cost across different LLM providers. Vanna also includes a feedback loop: users can upvote or downvote g