TRANSPARENCY · VERSION 1.1
Security & safeguards
This page describes the security controls currently implemented in the deployed chat foundation. It is updated alongside security-related releases. It is not a penetration-test certificate or a claim that all risks have been eliminated.
1. Current safeguards
- Password-gated access. The application requires a configured access password and refuses login if the password or session secret is below the configured minimum length.
- Signed, short-lived sessions. Session cookies are HttpOnly, Secure, SameSite=Strict and expire after 12 hours. Malformed, invalid, expired or implausibly long-lived signed payloads are rejected.
- Same-origin enforcement for login, logout, chat and provider diagnostics.
- Security response headers, including HTTP Strict Transport Security (HSTS) for one year, a restrictive Content Security Policy, frame protection, MIME sniffing protection, referrer policy and disabled camera, microphone and geolocation permissions.
- Request body, message length, message count and conversation-context limits.
- Bounded in-memory rate-limit storage with expiry and a maximum number of tracked keys.
- Rate-limit identity uses the socket peer address, not an unverified X-Forwarded-For value. Authenticated chat and provider diagnostics are keyed to the signed session.
- Gemini credentials remain on the server. The browser does not receive the API key.
- Application logs are designed to record model and token counts without storing chat prompts or replies.
- Pull-request CI checks server and browser syntax, audits production dependencies for high-severity vulnerabilities, and runs automated tests. GitHub Actions are pinned to reviewed commit SHAs. Deployment runs only after the test job succeeds.
2. Current limits and risks
A committed npm lockfile is not present yet. CI audits the resolved production dependency tree, but reproducible dependency installation remains a follow-up hardening task.
- The application currently uses one shared access password rather than individual user accounts, per-user roles or multi-factor authentication.
- Sessions are not stored server-side and there is no individual session-revocation list. Rotating the session secret invalidates all sessions.
- Rate limits are in-memory and apply per running application process. They are not a distributed abuse-control system or a guaranteed billing ceiling.
- The platform's trusted-proxy chain and client-IP forwarding contract have not been independently confirmed. Therefore, forwarded client-IP headers are not trusted for security decisions. The socket peer may be a shared proxy address, which can make login throttling affect multiple users behind that proxy.
- There is no persistent chat history, database, file upload, voice activation, research connector, background agent or autonomous external action in this foundation.
- No independent penetration test, formal privacy audit, incident-response exercise or disaster-recovery test has been completed.
3. Data and privacy boundary
Chat messages are sent to Google's Gemini API to produce replies. The application does not intentionally write prompts or replies to its own logs or persistent database. See the Privacy notice for provider processing, retention boundaries and user choices.
4. Security reporting
If you notice suspicious activity, unexpected access or a security weakness, stop using the service and contact the administrator through the channel used to provide access. Do not publish exploit details or send secrets in a report. A dedicated security contact must be published before public launch.
5. Release and update discipline
Security changes are developed on an isolated branch, reviewed against the existing implementation, tested in CI, merged only when required checks pass, and then verified on the live deployment. Material changes to data handling must update this page and the Privacy notice in the same release. Known limitations must remain visible rather than being described as completed controls.
6. Change history
- Version 1.1 — 21 November 2026: adds HSTS to the response-header baseline and corrects the daily request-limit wording to reflect session scope.
- Version 1.0 — 21 November 2026: first interface-visible security summary for the current chat foundation; documents session, origin, header, request-boundary, rate-limit and credential protections and their known limitations.
Version 1.1 · 21 November 2026 · Privacy notice · Return to Meghana