Source and model data flow
Authorized source snapshots are retrieved by a separate service, stripped of Git credentials, and supplied to audit workers. Only in-scope source and review context are sent to your configured model provider; excluded source content is withheld from models and execution. Provider contracts and configured retention determine their handling.
Access & artifacts
Membership roles are checked on the server for protected operations. Reports require authenticated organization membership. Private S3-compatible artifacts use organization-bound keys and protected single-use download tickets. TOTP multi-factor authentication and recent verification protect enrolled accounts; production privileged actions require enrollment by default.
Untrusted execution
Customer source, repository instructions, generated output, and artifacts cannot grant platform permissions. A separate execution service accepts signed, short-lived, scoped requests only after real host conformance passes. It uses gVisor, namespace isolation, read-only mounts and bounded scratch storage. Missing or stale conformance blocks execution.
Retention & deletion
Organization settings control retention periods. Operators must configure and run the retention worker and verify backup retention as part of deployment. Deletion queues hide requested data immediately, remove private object versions, and retain signed deletion tombstones for backup restoration. Restores stay quarantined until deletion replay and manager review. See operator runbooks for details.
Deployment responsibilities
TLS, encrypted storage, secret management, backup access, and the selected model-provider contract must be configured and checked before customer onboarding. No zero-retention or geographic-residency promise is made.