• Do not open a public issue for security vulnerabilities.
• Report privately to the repository maintainers (issue tracker's private security advisory, or the contact listed in the repository).
• Include: affected version, reproduction steps, potential impact and a suggested fix if possible.
• We aim to acknowledge reports within 7 days and publish fixes as soon as practical.

The optional password is a local privacy lock, not a security boundary:

• Client-side only: the gate (PasswordGate component) hides the UI in the browser. Backend API routes have no authentication — anyone who can reach the server over the network can call /api/* directly and read all data, bypassing the lock screen entirely.
• Storage: the password is stored as a hash in localStorage; unlock state lives in sessionStorage (cleared on tab close and on hard refresh).
• Hashing: a simple 32-bit non-cryptographic hash, no salt, no key stretching. It must NOT be considered resistant to brute force or dictionary attacks.
• No rate limiting: there is no lockout after failed attempts.
• Recovery: a forgotten password can only be reset by erasing all data (three-step confirmation + typing "DELETE ALL").

Therefore: keep the app bound to localhost (or a trusted network), and rely on OS-level access control for real protection. Do not treat the password as defense against anyone with network or filesystem access.

• API keys: stored only in your local SQLite database (data/ai-studio.db). Never share this file or commit it to version control (it is gitignored by default).
• Password protection: enable the optional lock screen in Settings to gate access to the app. Note it is a local convenience lock, not a substitute for OS-level access control.
• Network exposure: run the app on localhost unless you understand the risks of exposing it on a network; the API routes have no authentication of their own.
• Proxy settings: if you configure a proxy, ensure you trust it — all LLM traffic flows through it.
• Dependencies: run npm audit periodically and keep dependencies up to date.

• Conversations, settings and usage statistics never leave your machine except when you send a chat request to your own configured endpoint.
• Edge-TTS requests go to Microsoft's public TTS service with the text you choose to read aloud.