Vanna

Что такое Vanna?

Vanna is an AI SQL agent and agent-building framework for letting users query company data through natural language. The homepage positions it as a way to build agents that ship to production, with user-scoped execution, rich UI components, and quota management built in.

The product supports multiple LLM providers and databases, and the docs show a configuration flow that generates ready-to-use code for a chosen stack. Users can start with a sample notebook or self-hosted setup, then connect the framework to their own database, agent memory, authentication logic, and server runtime.

Что умеет Vanna?

Model and database selection

Vanna works with multiple LLM providers and databases. The homepage lists Anthropic, OpenAI, Gemini, and Ollama, along with SQLite, PostgreSQL, MySQL, Snowflake, BigQuery, and other databases.

Natural-language data access

The product supports natural-language querying over databases and is presented as an AI SQL agent that helps users ask questions without writing SQL directly.

Agent memory for reuse

Docs show agent memory that stores successful question-to-tool-argument examples so later queries can reuse prior interactions and improve accuracy.

User-scoped execution and access control

The configuration docs include user authentication and access groups, letting different users be assigned different permissions before the agent runs a tool.

Hosted admin features

The homepage describes optional hosted admin features such as observability, audit logs, file storage, data retention, and hosted agent memory.

Copy-paste configuration and server setup

The docs include ready-to-run server examples and generated configuration code, including FastAPI-based execution.

Варианты использования

“Self-serve database questions”

Let business or operations users ask questions against a company database in plain language instead of writing SQL manually.

“User-scoped internal agents”

Build a production agent that uses access groups, user resolution, and tool registration so different users see different capabilities.

“Prototype with sample or own data”

Start with the sample notebook or generated configuration code to test Vanna on SQLite or other supported databases before wiring it into a larger app.

“Operationalize agent usage”

Add hosted admin capabilities such as audit logs, observability, and data retention when a team needs more control over usage and records.

Часто задаваемые вопросы

How do you get started with Vanna?

Vanna’s docs show copy-paste-ready setup code for a chosen LLM provider and database. The quick start also offers a sample notebook on Google Colab to try Vanna with sample data.

What environments can Vanna run in?

The source shows Vanna working with multiple LLM providers and databases, and it supports self-hosting, Vanna Cloud, and enterprise deployment options.

Can you upgrade from open source to a paid plan later?

The pricing page says the tiers use the same core framework, so you can start self-hosting with the open source version and later move to managed cloud or enterprise deployment without rewriting code.

What does Vanna produce for users?

The docs and homepage show outputs that can include ready-to-use code, SQL execution, and data visualization through registered tools, depending on your configuration.