Vanna
Vanna is an AI-powered business intelligence assistant that connects to your Snowflake, BigQuery, or other database and helps you get insig...
Last verified:
What is Vanna?
Vanna 2.0 is a user-aware AI agent framework that connects LLMs to your database, enabling natural language querying of SQL databases. At its core, Vanna allows you to connect tools to an LLM (giving your AI agent the ability to run SQL queries, generate charts, or custom functionality), remember how to use them through Tool Memory that learns from successful interactions, and enforce permissions with user-specific access controls. The agent gets better over time by saving every successful interaction to Tool Memory, searching past SQL queries and tool usage patterns, and adapting successful examples to current context without manual training.
Key features include accurate Text-to-SQL generation via LLMs using Agentic Retrieval, multi-database support (Snowflake, PostgreSQL, MySQL, BigQuery, and more), multi-turn conversations, open-source availability, self-learning capabilities that improve with each interaction, and optional hosted admin features including access control, observability with real-time metrics and distributed tracing, agent memory with hosted vector database, secure file system, audit logs, and data retention. Vanna works with any LLM provider and can be deployed in Jupyter, Streamlit, Flask, Slack, Teams, and other frontends.
Vanna is designed for organizations wanting to empower users across their company—regardless of SQL expertise—to interact with databases using natural language. It's ideal for data analysts, product managers, business users, developers, and teams who need to democratize data access while maintaining security and access controls. The tool is built for production from day one with user permissions, quota management, and rich UI components requiring no rewrites.
Vanna pricing
Pricing model: Freemium
Vanna offers multiple pricing options. The open source version is 100% free to use with no cost per query, but users handle their own implementation and maintenance. For hosted Vanna Cloud: Free tier includes GPT-3.5 class LLM, daily token limit, hosted metadata retrieval, and Discord community support. Premium tier is $25/month with GPT-4 class LLM, 500 LLM requests per month, hosted metadata retrieval, and email support (extra 1000 requests for $50 monthly). For Vanna 2.0 managed platform: Explorer plan is $50/month for 2-3 people with 20 questions/day, admin features, API, and same-day email support. Team plan is $500/month for small companies with 300 questions/day, setup support, and same-day live support. Enterprise plan is custom pricing with unlimited questions, on-prem deployment support, SAML SSO, Admin API, advanced integrations, and custom work.
Vanna pros
- 100% open source and free to use
- Connects to any SQL database (Snowflake, PostgreSQL, MySQL, BigQuery)
- Works with any LLM provider
- High accuracy Text-to-SQL generation on complex datasets
- Self-learning improves performance with each interaction
- No manual training required—learns automatically from successful queries
- Multi-turn conversation support for complex queries
- User-aware access control and permissions enforcement
- Built for production with quota management from day one
- Easy frontend integrations (Jupyter, Streamlit, Slack, Teams, Flask)
- Hosted vector database for agent memory
- Real-time observability with metrics and distributed tracing
- Comprehensive audit logs for compliance
- Secure file system for file storage and retrieval
- On-premise deployment support for data privacy
- API available for custom integrations
- Tool Memory adapts successful examples to new contexts
- Supports multi-database environments simultaneously
Vanna cons
- Initial setup requires manual SQL samples for best results
- Documentation less plentiful than SaaS tools
- Technical setup complexity depending on integration level
- Performance varies based on chosen LLM
- May require fine-tuning for niche use cases
- Accuracy heavily depends on training data quality
- Learning curve to adapt to specific database structures
- Extremely complex queries may have limitations
- Free tier has daily token limits and rate limits
- GPT-3.5 in free tier less powerful than GPT-4
Frequently asked questions about Vanna
What is Vanna 2.0?
Vanna 2.0 is a user-aware AI agent framework that connects LLMs to your database. It lets you connect tools to an LLM (giving your AI agent ability to run SQL queries, generate charts, or custom functionality), remember how to use them through Tool Memory that learns from successful interactions, and enforce permissions with user-specific access controls that flow through the entire system.
How does Vanna get better over time?
Every successful interaction is saved to Tool Memory. When a user asks a similar question, the agent searches past SQL queries and tool usage patterns, then adapts successful examples to the current context. Your agent learns your database schema, business logic, and query patterns automatically without manual training.
What databases does Vanna support?
Vanna works with any database including Snowflake, PostgreSQL, MySQL, BigQuery, SQLite, and many more. It's a multi-database framework that you can configure to connect to your specific data source.
Which LLM providers work with Vanna?
Vanna works with any LLM provider. You can choose your language model from various options including GPT-3.5, GPT-4, and other LLMs. The open source framework allows you to select whatever LLM provider fits your needs.
How do I train Vanna?
The easiest way is through natural conversation in the chat interface: ask a question, review the response, and provide feedback if it needs improvement. The agent automatically learns from successful corrections. You can also manually add question-SQL pairs and business context documentation to agent memory for structured training.
What admin features does Vanna provide?
Vanna provides optional hosted admin features including Access Control (control who sees what), Observability (real-time metrics and distributed tracing), Agent Memory (hosted vector database for learning), File System (secure file storage and retrieval), Audit Logs (comprehensive activity tracking and compliance), and Data Retention (long-term storage for conversations and logs).
Is Vanna open source?
Yes, Vanna is 100% open source. You can start self-hosting with the open source version at no cost, then seamlessly upgrade to the managed cloud platform if needed. The open source framework works with any LLM provider and database.
Can I deploy Vanna on-premise?
Yes, Vanna supports on-premise deployment, especially in the Enterprise plan. This satisfies various data privacy challenges and allows secure, private hosting for organizations with strict data security requirements.
What frontend integrations does Vanna support?
Vanna can be deployed using Jupyter Notebook, web apps, Slackbot, Streamlit apps, Flask, Teams, and any other platform. It has simple integrations into many UIs with easy UI deployment options.
How does access control work in Vanna?
Every user has specific access controls that flow through the entire system. Access control uses group-based membership (not user-based) for easier management, follows the principle of least privilege (only grant access when needed), and empty lists mean accessible to all. Custom features should be documented and tested with different user types to verify access controls work properly.