Trust & security
Agent 99 is built on the Chibitek Stiki security control plane. This page describes controls that are implemented and verifiable today — not a roadmap.
Identity
All authentication through Chibitek Stiki SSO (auth.chibitek.com) — RS256 JWTs verified against JWKS, platform grants enforced fail-closed in middleware before anything renders. No local passwords, ever.
Tenant isolation
Each client organization runs in its own database, resolved only from the signed org claim. Row-level security applies inside every tenant database as defense in depth. Unmapped orgs get nothing.
Audit trail
Every mutation writes an append-only activity event (actor, action, risk, outcome). Incident timelines are append-only: reversals are new events, never edits. Platform audit streams to Stiki.
AI governance
Agents start in shadow mode. Execution requires human approval; actions are bounded (acknowledge, escalate, note, resolve-recovered) and reversible. Trust scores move with your decisions. Kill any agent by demoting it to shadow.
Secrets
API tokens are bcrypt-hashed at rest and shown once. Ingest and webhook deliveries are HMAC-SHA256-signed. Vendor credentials live in the Stiki broker — never in this platform.
Transport & storage
TLS everywhere; managed Postgres (Supabase) with encryption at rest; soft deletes preserve the forensic record.
Command governance
Nova Shell only ever suggests from a policy allowlist; every dry-run is audited with its risk level. No browser-side execution exists.
Development lifecycle
Every change ships via pull request with typecheck, lint, build, and fail-closed smoke tests in CI. Version governance ties every release to a changelog entry.
Security questions or reports: security@chibitek.com.