Offline pre-transaction security for AI agents and server-side wallet APIs.
We build deterministic guardrails that sit after policy/simulation and immediately before the signing boundary (MPC/HSM). The goal is to prevent “what you see ≠ what gets signed” failures—address poisoning, alias confusables, and last-mile drift—by validating the canonical payload right before execution.
SealGuard (beta) — a stateless pre-sign validator that canonicalizes the transaction payload and returns an auditable verdict (ok / step-up / block) with deterministic reason codes.
TxSeal — emits DSSE receipts for signing decisions and evidence so flows can be verified offline. Includes a zero-runtime-deps DSSE+JCS verifier.
PoisonAtlas — dataset format for address-poisoning research and evaluation.
Private beta. The core engine and full rule sets are iterated with a small set of design partners. Expect breaking changes until a 1.x GA release is announced.
Security reports: please follow SECURITY.md in this org’s .github repository.