How we protect LOVA — and you
AI Privacy
& Security
Every conversation with LOVA passes through eight layers of defense. No training on your data. No exposed prompts. No autonomous money movement.
8 Layers of AI Defense
Each layer is independently enforced. Bypassing one would not compromise the others.
Isolated AI Gateway
LOVA runs through Lovable AI Gateway with no-training agreements. Your conversations are never used to train Google Gemini or OpenAI models.
Server-Side Prompts
All LOVA system prompts and persona instructions live in encrypted Edge Functions — never exposed to the browser. Hackers can't read or modify her 'brain' from DevTools.
Row-Level Security (RLS)
Every database table enforces RLS — LOVA can only ever read data belonging to the currently authenticated user. Cross-user leaks are mathematically impossible.
Zero Key Exposure
LOVA never sees your seed phrase, private keys, PIN, or password. These are encrypted locally with vaultCrypto and never leave your device.
Tool-Calling Sandbox
LOVA returns structured JSON via tool-calling schemas. She physically cannot 'escape' the schema or trigger arbitrary actions — even with a prompt injection attempt.
No Autonomous Transactions
LOVA cannot move LOVE, sign transactions, or change settings. Every financial action requires your explicit PIN or biometric confirmation through a separate Edge Function.
Stateless by Design
Each AI request is independent. Conversation history stays in your account only — the model itself remembers nothing between sessions.
Rate Limit & Abuse Shield
Per-user rate limits and 429/402 handling block DDoS, scraping, and credit-burning attacks at the gateway layer.
Real Threats. Real Defenses.
We don't pretend AI is risk-free. Here's what could go wrong — and exactly how each scenario is contained.
What LOVA Never Does
Hard guarantees. Enforced by architecture, not promises.
Trust, Verified by Architecture
Read the full technical details in our Whitepaper, or reach out if you want to audit a specific layer.
