Vanna AI
Vanna 2.0 enables organizations to build multi-user AI agents that query databases using natural language, eliminating the need to write SQL manually. Built with identity-first architecture, it handles user permissions, access control, and quota management natively—making it production-ready without additional wrapper frameworks. The framework supports any LLM provider and database, from SQLite to Snowflake, and includes optional hosted features like audit logs, observability, and secure file storage for enterprise deployments.
Vanna 2.0 is an open-source agent framework that enables organizations to build multi-user AI applications for querying databases and other systems using natural language, eliminating manual SQL writing. Built on an identity-first architecture, it natively handles user permissions, access control, quota management, and multi-tenant isolation—making it production-ready without requiring additional wrapper frameworks or custom security layers. The framework supports any LLM provider (Anthropic, OpenAI, Gemini, Ollama) and database (SQLite, PostgreSQL, MySQL, Snowflake, BigQuery, and others), with a dual-output model that returns concise summaries for LLM reasoning alongside rich interactive components for users, such as sortable data tables with export functionality. Beyond SQL querying, Vanna 2.0 powers coding assistants, document Q&A with document-level permissions, workflow automation, and customer support tools—all with user-scoped execution and quota enforcement built in. Optional hosted features include audit logs, observability with real-time metrics and distributed tracing, a hosted vector database for agent memory, secure file storage, and conversation persistence with encryption. The framework is MIT-licensed with no vendor lock-in, supports both self-hosted and managed deployment models, and is used by organizations including Apple, NVIDIA, Tesla, and Samsung. Developers can prototype with MockLlmService in minutes, then scale to production by adding real LLMs, enabling streaming, and deploying without architectural rewrites.
The privacy policy does not explicitly state whether user content is used to train AI models. However, the broad use of
No training opt-out mechanism was identified in the legal documents.
High risk across multiple dimensions. Only use with public, non-sensitive data.
The privacy policy does not explicitly state whether user content is used to train AI models. However, the broad use of information for 'Analytics and Improvement' and the collection of 'Communications' and 'Usage Data' creates moderate risk that this data could be used for model training without explicit consent.
“We collect information that you voluntarily provide when using our Services, including: Communications: Feedback, survey responses, support inquiries, and other communications you send to us”
The Terms of Service restrict user redistribution of site content but do not clearly define what license users grant for their own uploaded content or data. The broad sharing provisions create moderate ambiguity about content ownership and usage rights.
“You may not: Reproduce or distribute the materials to third parties without prior written permission”
The policy explicitly shares data with multiple third parties including Service Providers, Business Partners, Affiliates, and for Legal Requirements. Third-party analytics services (Google Analytics, HubSpot, AWS Analytics) are named, creating significant third-party sharing risk.
“We use third-party analytics services: Google Analytics, HubSpot, AWS Analytics, Pixel Tags. How We Share Your Information: Service Providers, Business Partners, Affiliates, Legal Requirements, Business Transfers”
The data retention policy is vague, stating data is retained 'as long as necessary' without specific timeframes or clear deletion procedures. Users have limited visibility into how long their data persists or how to request deletion.
“We retain personal information as long as necessary to provide Services, comply with legal obligations, resolve disputes, and maintain business records.”
The terms do not explicitly address ownership of user-generated content or AI outputs. The site content is protected by Vanna.AI's IP, but user content ownership is not clearly defined, creating moderate ambiguity.
“The content on this Site, including but not limited to text, software, code, designs, graphics, photos, sounds, music, videos, and other materials, is protected by copyright, trademark, and other intellectual property laws.”
The policy mentions CCPA compliance for California residents and references industry-standard security measures, but does not explicitly claim SOC2, GDPR, HIPAA, or ISO certifications. Compliance is partial and not comprehensively documented.
“We implement industry-standard security measures including: Encryption of data in transit and at rest, Regular security assessments and audits, Access controls and authentication mechanisms”
Track and measure your AI tools with Olakai's observability platform.
Explore Olakai