Security

Security and reliability

What we guarantee in code, how OhAuthCat runs today, and what is still planned.

  • Sign-in rulesBuilt and tested12 of 12Each is a product requirement with a named test.
  • Regions liveap-south-11Mumbai, across availability zones, behind Cloudflare.
  • Reliability workPlanned2 plannedStandby path, backups and more regions, listed below.
  • CertificationsNot certifiedNone yetWe say so here until that changes.

We hold no security certifications yet, and we run in one AWS region. This page says so until that changes.

Rules every sign‑in follows

These are product requirements with named tests. A check is never weakened to make something work.

  • Sign‑in codes work onceBuilt and testedAuthorization codes expire after 60 seconds and can be exchanged only once. A second attempt fails and revokes what the first one issued.
  • Your secret key is always requiredBuilt and testedExchanging a code always needs your secret API key and must match the environment, organization and redirect URI it was issued for. PKCE is checked on top, never instead.
  • Exact redirect URIsBuilt and testedRedirects go only to URIs on your allow-list, matched exactly. No wildcards in production.
  • Only verified domains sign inBuilt and testedA user is signed in only if their email domain belongs to a domain the organization has proven it owns. A misconfigured identity provider for one company can’t sign in people from another.
  • Strict SAML handlingBuilt and testedSize limits, no DOCTYPEs or external entities, one parse, and identity read only from the element the signature actually covers. That closes signature-wrapping and parser-differential attacks. Exactly one assertion, SHA-256 or stronger by default.
  • Replays are refusedBuilt and testedResponses must answer the request we sent, and that request’s state can be used once. Every time, audience, destination and issuer condition is checked.
  • Outages never weaken checksBuilt and testedIf a database, cache or key service is unavailable, sign-ins may fail. A check is never skipped to keep them working.
  • SCIM tokens are stored hashedBuilt and testedDirectory tokens are 256-bit random values, shown once and stored only as hashes. Rotating one keeps the same directory and endpoint.
  • Safe fetching of your URLsBuilt and testedMetadata, discovery and webhook URLs are fetched over https only, never to private or cloud-metadata addresses, with every redirect re-checked.
  • Secrets are encrypted and never shown againBuilt and testedSecrets are envelope-encrypted with keys held in a cloud key service, never logged and never returned after creation.
  • Tenant isolation in the data layerBuilt and testedEvery database query is scoped to an environment by type, and tests try cross-tenant access on every endpoint.
  • Phishing-resistant dashboard sign‑inBuilt and testedMFA on every plan with passkeys preferred, and backup passkeys and recovery codes so an outage of our own sign-in never locks you out.

Where and how it runs

  • One AWS region: Mumbai (ap-south-1)Built and testedThe API, sign-in and SCIM run in a single AWS region across availability zones, behind Cloudflare. The websites and dashboards are static and served globally from a CDN.
  • Standby path if Cloudflare is downPlannedA second domain, oauthcat.com, is hosted on AWS Route 53 without Cloudflare. Our service-provider metadata is built to list its sign-in URL, so IT admins can register it once at setup. The standby load balancer itself is provisioned when there is traffic to justify it; until then the standby hosts have no DNS records and our SDKs fall back to the primary error.
  • Backups and region recoveryBuilt and testedThe database keeps point-in-time recovery for 7 days, and encrypted backups are taken every hour and every 6 hours, with copies kept in a second Indian region (Hyderabad, ap-south-2) and retained for 60 days. Losing the whole primary region means a rebuild there from code and backups, measured in hours, not seconds; that rebuild has not been rehearsed yet.
  • Multiple regionsPlannedThe code is built to run in several regions and is tested that way locally. We will only say we run multi-region when we actually do.

We engineer sign-in and SCIM for 99.95% availability in this single-region setup. That is an internal target, not a contractual SLA.

Your data

  • We store the configuration you create, sign-in outcomes, SCIM requests with values removed, and audit logs. Retention depends on your plan, and old records are deleted automatically.
  • We never log tokens, secrets, SAML payloads or full email addresses.
  • Audit logs are append-only: not even an account owner can edit or delete an entry.
  • Support access is read-only, needs a reason and is written to your audit log. Staff tools are not reachable from the public internet.
  • Data is processed in AWS ap-south-1 (Mumbai, India).

More

Sub-processors and policies are in the trust centre. To report a vulnerability, follow our vulnerability disclosure policy.