How Vana Works

Today's AI is built on data that users don't control or benefit from, leaving valuable private datasets locked away in corporate silos. Users generate enormous amounts of data but have no ownership stake in the AI systems trained on it.

Vana solves this through two core technical innovations that enable user-owned AI:

VRC-20 Data Tokens and DLPs: Programmable Data Ownership

VRC-20 tokens make data ownership liquid and programmable. Unlike regular tokens, VRC-20s are cryptographically bound to specific datasets through proof of contribution mechanisms. When users contribute data to a data liquidity pool (DLP), they receive VRC-20 tokens representing their ownership stake. These tokens are:

• Earned through contribution: Users receive tokens by providing proof-of-contribution validated data to Data Liquidity Pools
• Burned for access: AI builders must burn both VANA and VRC-20 tokens to access datasets
• Fully programmable: ERC-20 compatible, enabling complex economic logic and DeFi integration
• Attribution-preserving: Each token traces back to specific data contributions and evm-compatible wallet addresses

This creates liquid markets for data assets while maintaining granular attribution to individual contributors.

Secure Runtime with Granular Enforced Permissions

The secure compute layer enables computation on private data while enforcing user-specified permissions. Built on Trusted Execution Environments (TEEs), the runtime:

• Enforces access controls: Only executes on data when proper tokens are burned and permissions granted
• Preserves privacy: Raw data never leaves the encrypted environment during computation
• Enables granular permissions: Users can specify exactly what data fields can be accessed via specific SQL queries on their structure data
• Provides verifiable execution: All compute jobs require blockchain approval and results are written onchain

For example, a researcher can access users' health history to train a recommendation model, but the TEE ensures they never see individual user identities and that raw data never leaves the secure environment.

How They Work Together

When users contribute data, they receive VRC-20 tokens representing ownership. When AI builders want to train models, they burn these tokens to gain access. The secure runtime ensures that even with access, builders can only use data according to the specific permissions users granted.

This creates a unified system where:

• Users retain ownership and earn from their data's use in AI training
• AI builders can access valuable datasets previously locked in silos
• Privacy is preserved through cryptographic enforcement rather than trust
• Economic value flows directly back to data contributors

The result is user-owned AI: models trained on private data where users control access and benefit from the value created.