TL;DR: AI models have exhausted public data. The next performance leap requires private datasets like enterprise records, verified transaction histories, and app usage logs. Vana and Brevis eliminate the three blockers preventing this: Brevis zkTLS proves data authenticity from any source (banks, exchanges, apps) without exposure. Pico zkVM computes on that data locally and proves correctness without revealing raw information. Vana coordinates user consent and rewards. Developers get verified, high-fidelity training data. Users retain control and get paid.
AI’s Data Problem
AI models trained on public data alone have plateaued. Web scrapes and data broker feeds provided the initial training corpus, but this approach has hit diminishing returns. The next breakthrough requires private, high-fidelity data: verified enterprise datasets, financial transaction histories, app usage logs, and user-generated insights. This data offers higher quality and better context than public datasets.
Three problems block access to private data at scale:
Incentives: Users need fair compensation for contributing data.
Privacy: Users must control what they reveal. Proving trading volume shouldn’t require exposing every individual trade.
Verifiability: Developers must verify data origin and computation correctness without seeing raw data.
Vana and Brevis solve all three.
How Brevis and Vana Work Together
Vana operates an open network for user-owned data. Users contribute information to Data Collectives in exchange for rewards or tokenize their data directly. Users retain ownership while creating a market for verified, consented data.
Brevis provides the cryptographic layer that makes this trustless:
zkTLS Coprocessor: Proves data authenticity from any source (banks, exchanges, web apps) without exposing the data itself. A user can prove “this transaction history came from Coinbase” without revealing which trades occurred.
Pico zkVM: Processes verified data locally, computing aggregates, averages, or filtered metrics. Then generates a zero-knowledge proof of correctness. The computation runs on the user’s machine. Only the proven result goes on-chain.
The pipeline works like this:
- zkTLS cryptographically proves data origin from the source
- User computes locally on verified data
- Pico zkVM generates proof of correct computation
- Only the final metric and proof become visible
- Vana coordinates consent and distributes rewards
Developers access verified insights without intermediaries or privacy compromise. Users monetize information without surrendering control.
Concrete Example: AI Trading Assistant
A Vana Data Collective powers an AI trading assistant that needs verified volume data. The model doesn’t need full transaction histories, just proven metrics like “this user traded $120K on ETH pairs.”
With Brevis:
- User authenticates their exchange data via zkTLS
- Computes token-specific volumes locally on their machine
- Generates a ZK proof confirming both data origin and calculation
- Reveals only the final metric: “$120K ETH trading volume”
The user’s complete trading history might total 10MB of transaction data. The ZK proof compresses this to a 200-byte verification that confirms the aggregate without exposing individual trades, counterparties, timestamps, or price levels. The AI receives high-integrity input without accessing sensitive information.
The proof verifies on-chain in under 5 seconds with minimal gas costs. The AI model gains training signal it can trust. The user gets compensated without compromising privacy.
What This Enables
Data Collectives can now offer verified datasets for AI training across any domain:
- Financial models trained on proven transaction patterns without accessing individual trades
- Healthcare AI using verified symptom correlations without patient identity exposure
- Recommendation systems built on authenticated usage data with selective disclosure
- Fraud detection trained on cryptographically verified behavioral patterns
The common pattern: AI models get high-fidelity, verifiable training data. Users control exactly what gets revealed and earn compensation for participation.
Get Started
Documentation:
Community: Join us on Discord, Telegram, or X (@brevis_zk) for support and discussion.
Vana Data Collectives powered by Brevis zkTLS and Pico zkVM turn private data into verifiable, monetizable assets without compromising user privacy.
Brevis Network© 2023 