Document and code review
What you can do.
Redact personal details, review contract terms, and check code for risks. Download the results and a record of the review.
- Personal data redaction
- Contract and code review
- Result and receipt downloads
What’s been built
- One place to review documents and codeReady
- Downloadable results and review recordsReady
- Paddle payment servicesPaddle approved
Technology & architectureMask sensitive details. Keep the review record.Explore the architecture
How it works, in three steps
Choose personal-data masking, document review, or code review in one workspace. Download the result and check which policy and model were used.
NFKC normalization precedes email and explicit API-key matching. Korean identifier date and checksum checks, plus the Luhn card check, determine the masking scope.
Contract facts come from deterministic extraction, with model output treated as advisory. Code review combines high-confidence rule checks with supplementary local model review.
The result and model metadata are hashed together and linked to the previous receipt hash. This work record is distinct from a source archive or a Groth16 computation proof.
NFKC normalization precedes email and explicit API-key matching. Korean identifier date and checksum checks, plus the Luhn card check, determine the masking scope.
Security & design principles
Encrypt for a checked environment
Azure MAA · X25519 · AES-GCMA separate masking SDK verifies Azure evidence before encrypting input. Ordinary document review follows a path without hardware attestation.
Keys and evidence bound to a request
Request binding · Replay protectionThe attested key, request, and tenant are bound together with expiry and replay checks. Verification also covers Azure signatures, VM identity, and the request-bound vTPM quote.
Computation proofs with a defined scope
Groth16 · Aptos MoveA separate engine proves an incorrect claimed sum of four integers from 25 to 99. Current records use Aptos localnet, TestUSD, and a single-party setup.