Loops Houseadytus

NEAR HACKATHON
Team
1 member- STOwner
steady
Overview
Problem: Companies want AI for their most sensitive work, like sales numbers, market data and board reports, but every mainstream AI tool sends that data in plaintext to a provider they have to trust. “We don’t train on your data” is a promise, not a proof.
What we built: adytus is a private AI workspace. Upload confidential files, and it analyzes them, runs calculations, builds charts and produces deliverables like a BCG-style PowerPoint slide. Chats and files are encrypted with the user’s own key, so even we can’t read what’s stored.
How NEAR fits:
Private inference on NEAR AI Cloud: open-weight models run inside Intel TDX + NVIDIA confidential-computing TEEs, so the host can’t read prompts. Verifiable proof for every answer: we check NEAR AI’s per-answer signature against the exact request and response bytes, and confirm the signer is the key in a fresh TDX attestation. Each reply shows “✓ Verified in TEE” with a public proof page, plus a CLI (go run ./cmd/verify) so anyone can check it themselves. Our own NEAR contract (adytus-pay.testnet) for top-ups in NEAR or USDC, plus login with a NEAR wallet (Meteor / MyNearWallet, sign-a-message, no fee).
adytus existed before the hackathon. The NEAR AI integration, signed receipts, verify page, contract and NEAR login were all built during the event (see the README).