Pre-launch · no production tenants yet
Trust is an architecture, not an attestation we bought.
UrvaIQ does not hold SOC 2 or ISO 27001 certification. We would rather show you exactly what we built than wave a badge. This page describes the architecture you can verify — every claim here is anchored to code, tests and public documentation.
Isolation architecture
- OPA policy decision point in an isolated runtime with no network route to your systems of record; it holds decisions, nothing else.
- Verification runs on a separate read-only identity with a restrictive egress allowlist. Your systems cannot be written to by the verifier.
- Ed25519 receipt signing lives in a dedicated signer — no database access, no audit access, keys in KMS with a published verification key set and rotation.
- Token vault: reversible tokenization in its own runtime with its own mount, dedicated credential, and RLS on every tenant-scoped table.
Data handling
- Evidence and receipts are append-only and hash-chained; deletions are blocked by database triggers, not by application courtesy.
- Sensitive originals live in vaulted tokenization only; routine logs carry identifiers, states, counts and hashes — never values.
- Legal hold overrides the retention purge; tenant data is isolated with FORCE RLS on every table in the set.
Development posture
- 69 published API operations against a versioned OpenAPI contract, with automated surface, schema and reason-code allowlist gates in CI.
- Fail-closed design reviewed per dependency (policy engine, IdP, connectors, queue, signer): a down dependency denies rather than serves.
- Full threat model (`docs/security/threat-model.md`) and a ten-row automated failure matrix.
Subprocessors
Pre-launch. When a pilot designates a deployment target, the subprocessor list, data-processing agreement and region are published here before any tenant data is processed. No third-party SDK telemeters user data from this site.
Responsible disclosure
Found a flaw? Report it to security@urvaiq.com. We acknowledge within two business days, do not threaten researchers, and credit valid reports unless you prefer anonymity.