Vanna 2.0

Vanna 2.0 is a framework that connects LLMs to databases to create AI agents capable of running SQL queries, generating charts, or performing custom functions. It is designed to support users across an organization, including those without SQL expertise, by enabling natural language interaction with data and enforcing permissions throughout workflows. Tool memory helps agents learn from successful interactions to improve over time, and memory adapts to the database schema and business logic without manual training.

Vanna 2.0 emphasizes remembering how to use tools by saving successful interactions to Tool Memory. When a user asks a similar question, the agent searches past SQL queries and tool usage patterns and adapts examples to the current context. It aims to provide fast adaptation from memory for similar questions and deeper investigation paths, while tracking and learning from previous runs.

The core model enables tool integration with an LLM, allowing agents to run SQL, generate visuals, or call custom functions. It enforces permissions so that each user’s access rules flow through the system, and it supports multiple databases and LLM providers. The framework is presented as an open source option with hosted admin features for observability and control.

Vanna 2.0 offers hosted admin features such as access control, observability, agent memory, file storage, audit logs, and long-term data retention. It also supports various databases and LLM providers, and provides a path to cloud or self-hosted deployments. Documentation and quickstart guides describe how to connect, configure, and migrate from earlier versions to 2.0.

Use cases

• Enable users to interact with a database using natural language, lowering SQL barriers and increasing data accessibility
• Provide controlled access to data with per-user permissions