Deployment topology
CelinQ Server runs entirely on infrastructure your organisation controls — on-premises or in your own cloud account. There is no CelinQ-operated service in the path of ordinary collaboration: the server your architects' local clients synchronise with is the one you deployed, and nothing about core sync or Fusion merging depends on an external CelinQ service being reachable.
Authentication
Two mechanisms, active side by side, for different purposes. Personal access tokens — PBKDF2-SHA256 hashed, shown once at issue, never stored or logged in plaintext — are meant for API access and service automation. Microsoft Entra ID SSO (standard OIDC, Authorization Code + PKCE) is available for human sign-in, with Entra roles or groups mapping, promote-only and licence-capped, onto CelinQ's own Viewer / Editor / Administrator / Owner roles. A non-Entra generic OIDC provider — Keycloak, Okta — is not yet implemented.
Transport encryption
All traffic between CelinQ Connect (the local companion service) and the CelinQ Server runs over TLS. Certificate pinning is available and opt-in on the client side — a SHA-256 fingerprint match configured on the client, an extra check against a specific certificate rather than trusting the certificate chain alone. It is client-side pinning only; there is no server-side pinning of client certificates in the current implementation.
Secrets storage
API keys and other server-side secrets — an AI provider's API key, for example —
are never stored in plaintext. On Windows, the server uses DPAPI at machine scope. On
other platforms, it uses AES-GCM with a 32-byte key injected via the
CELINQ_SECRET_KEY environment variable. If neither mechanism is
available, CelinQ refuses to persist the secret in plaintext rather than falling back
to an insecure default.
Backup
SQLite deployments have a real, tested online backup path — triggered through the
Control Plane or the admin API, verified in CelinQ's own test suite. PostgreSQL
deployments do not yet have an automated online backup built in; the server returns
an explicit error with guidance to use pg_dump or your database's own WAL
archiving instead, rather than silently doing nothing.
Authorization
Workspace roles — Viewer, Editor, Administrator, Owner — are enforced server-side on every call, not just checked once at login. A revoked or downgraded role takes effect on the next request, not the next sign-in.
API surface
gRPC for sync traffic; a versioned REST API (/api/v1) for the Control
Plane and admin operations. CelinQ has no built-in API gateway, rate limiting, or WAF
— the documented guidance is to deploy it behind your own reverse proxy or API
gateway when it is exposed beyond a trusted network.
What's not built yet, stated plainly. There is no published vulnerability-disclosure process for CelinQ today — no dedicated reporting address, no documented response process. If you find a security issue, reach us through the contact form and we will respond directly; a formal disclosure process is on the list of things worth building properly rather than announcing before it exists. Rate limiting and an API gateway are likewise not built in — deploy CelinQ behind your own.
Read next
Compatibility
Versions tested, formats supported, and exact cross-tool mapping status.
Data Residency for Architecture
Why where the model lives is not a detail for public-sector work.
Sovereign AI Deployment Patterns
What "no external AI exposure" actually means as a deploy-time setting.