Security Checklist

Use this checklist before exposing the deployment to users. Items describe operator responsibilities; the supplied Compose or Kubernetes files complete none of them automatically.

Secrets and configuration

• Replace every default in SECRET_KEY, POSTGRES_PASSWORD, REDIS_PASSWORD, and QDRANT_API_KEY.
• Set a unique non-empty INTERNAL_API_TOKEN shared by the API, worker, and sandbox-worker for the internal upload gateway.
• Keep .env, Kubernetes Secret manifests, and installer output out of version control with tight file permissions.
• Store production secrets in a secret manager or externally managed Secrets where possible; rotate during planned maintenance windows.
• Point API_BASE_URL/API_INTERNAL_BASE_URL at internal service URLs, and PUBLIC_API_URL, FRONTEND_URL, BACKEND_CORS_ORIGINS at the public browser origin.

Authentication and access

• Register the first account only while the deployment is still private; the first registered user becomes the initial superuser.
• Configure session timeout, registration, email/SSO, and other site settings in the admin UI per policy.
• Review team membership and admin permissions regularly; remove unused API keys and integrations.
• Before enabling MFA/SSO, test the provider configuration and recovery path.

Network and proxy

• Terminate HTTPS at the external reverse proxy or Ingress with valid certificates.
• Expose only the frontend/reverse-proxy port; keep PostgreSQL, Redis, Qdrant, and the direct API port on private networks.
• Set exact CORS origins; never use * in production.
• Configure streaming over HTTP/1.1 with proxy_buffering off and read/write timeouts matching the 1800-second example.
• Restrict the clouisle namespace with RBAC, NetworkPolicies, and Ingress policy.

Data protection and recovery

• Enable encryption at rest for PostgreSQL, Qdrant, uploads, and backup storage where supported.
• Back up PostgreSQL, Qdrant snapshots, and uploads on schedule; Redis holds only cache/queue state so backup is optional.
• Keep encrypted off-site copies with a retention policy.
• Rehearse a full restore including Qdrant and uploads; record RTO/RPO results.
• Never run docker compose down -v or delete PVCs outside intentional destructive operations.

Runtime and sandbox

• Pin production image tags/digests instead of latest; scan images before deploying.
• Keep sandbox filesystem isolation enabled with the worker's required seccomp/capability settings; review the sandbox hardening guidance before changing them.
• Limit sandbox egress and resource usage at the host/cluster layer per the threat model.
• Keep Docker, Kubernetes, PostgreSQL, Redis, Qdrant, Bubblewrap, and the host kernel patched.

Logging and incident response

• Collect and protect API, worker, beat, frontend, and proxy logs; exclude tokens and passwords.
• Poll GET /api/v1/health and use the admin observability endpoints/dashboard (requires admin:dashboard:access). There is no built-in Prometheus /metrics endpoint.
• Alert on failed health checks, auth anomalies, queue backlog, disk exhaustion, and backup failures.
• Document escalation, credential rotation, isolation, restore, and post-incident review procedures.