Security

Last updated: October 7, 2026

Overdo serves open-weight models from our own inference infrastructure. That architecture choice is also our biggest security and privacy control — here is exactly what we do, and what we don't.

Your prompts stay ours

  • Chat, code, vision, image, and video requests run on open-weight models served from our own GPU infrastructure. Your prompts are not sent to OpenAI, Anthropic, Google, or any other third-party model provider.
  • No third-party analytics or trackers. Product metrics are first-party events we record ourselves — no GA, no Segment, no ad pixels.
  • Payment card details never touch our servers — checkout and card storage are handled entirely by Stripe.

Accounts and sessions

  • Passwords are hashed with scrypt and a per-user salt — we never store them in plaintext.
  • Sessions are HMAC-signed tokens in httpOnly, Secure cookies; sign-in is verified server-side with timing-safe comparison.
  • Sign-in, sign-up, and verification endpoints are rate-limited against brute force.

Data in transit and on the box

  • TLS on every endpoint — web, API, media, and inference traffic — with HSTS enabled.
  • Outbound fetching (web search, page reader) is hardened against SSRF: private and link-local network ranges are unreachable from our fetchers.
  • Conversations and usage records are stored on our own servers. We don't sell or share them — the business model is subscriptions and API, not data.

What we don't claim

We're an early-stage platform, not a certified enterprise. We do not currently hold SOC 2 or ISO 27001 certification, and we don't offer SSO/SAML or per-customer key scoping yet. Enterprise-grade controls are on the roadmap — if your organization needs them, talk to us about what that looks like.

Vulnerability reporting

Found a security issue? We appreciate responsible disclosure. Report vulnerabilities to [email protected] — or raise a request in the product (chat → Need help? — sign-in required). Include steps to reproduce and allow us reasonable time to remediate before public disclosure.