Loops HouseSafeRedline

NEAR HACKATHON
Links
Team
1 member- ALOwner
Alex Maldonado
Overview
An NDA is the one document you can't paste into a chatbot. It's confidential, often under NDA itself, and a normal AI service gives you a privacy policy, not proof.
SafeRedline is a contract and NDA reviewer you run on your own computer. You load a contract. It flags the risky clauses, quotes them word for word, explains the trap in plain English, and drafts a fairer replacement you can paste back. SafeRedline stores nothing.
The AI runs on NEAR AI Cloud, inside a hardware "locked room" (a TEE: Intel TDX plus NVIDIA confidential GPUs). The part I care about most is that you can check the lock.
Before your contract leaves your machine, SafeRedline asks the room to prove itself, using a fresh random number so an old proof can't be replayed. It checks Intel's and NVIDIA’s signed evidence, confirms the connection is pinned to the key inside that room, and only then sends the text. When the answer comes back, it checks the room's signature on the exact words sent and received.
If any check fails, nothing is sent. Missing evidence counts as a failure too. Live, every counted check passed (23 before sending, 29 with the signed answer), and 100+ offline tests plus a simulated twin cover the failure paths.
What it can't prove: the enclave software is identified by its hash, not audited; the NEAR gateway and model still see your text inside the room; and my own local server isn't attested.
NEAR makes this possible. Without NEAR AI Cloud, this is a chatbot with a promise. With it, privacy is something you can check.
Other AI contract tools promise zero retention. SafeRedline proves where your contract ran.