Privacy Layer

Privacy Layer

Privacy-preserving operations for AI agents handling sensitive financial data.

ZK-SNARK Private Transfers (Railgun Protocol)

Railgun enables fully private ERC-20 transfers using ZK-SNARKs on Ethereum. Three core operations:

Shield (Public → Private)

Move tokens from a public address into the Railgun privacy pool:

• Tokens become shielded — balance hidden from chain observers
• Agent receives a private note commitment
• On-chain: only the deposit amount and token are visible

Private Transfer (Private → Private)

Transfer tokens within the privacy pool:

• Sender and receiver are hidden on-chain
• Amount is hidden via ZK proof
• Only the proof validity is verifiable

Unshield (Private → Public)

Withdraw tokens from the privacy pool to a public address:

• Converts shielded balance back to standard ERC-20
• Receiver address becomes public

Railgun Contract Addresses

Network Egress Monitoring

Monitor all outbound connections from the agent to detect data leaks:

• Track all RPC calls, HTTP requests, and WebSocket connections
• Classify targets as external vs loopback
• Log destination hosts, ports, and protocols
• Alert on unexpected outbound connections

Agent Privacy Architecture

1. Secret Isolation

• Private keys never sent to LLM — all signing handled in isolated process
• Mnemonic stored with 0600 permissions (owner-only read)
• Config directory 0700 permissions

2. LLM Security Fence

• System prompt forbids revealing private keys, mnemonics, API keys
• Reduced tool surface — only expose necessary operations
• ClawWall DLP catches secrets before they reach the LLM

3. Private Inference

• Use Venice AI or local Ollama for zero-retention inference
• Sensitive data (portfolio values, strategies) processed without provider storage
• No training on agent queries

4. On-Chain Privacy

• Shield balances via Railgun before agent operations
• Private transfers between agent sub-wallets
• Unshield only when public settlement is needed

Privacy Checklist for Agents

Usage Tips

• Shield tokens before any sensitive agent operations
• Use private transfers for inter-agent payments
• Monitor network egress during development to catch unexpected leaks
• Always use Venice or local models for strategy-sensitive reasoning
• Never expose raw portfolio data to third-party APIs without shielding