Please report security problems privately, not in a public issue:

• through GitHub: Report a vulnerability, or
• by email to security@archdraw.dev.

Include what you found, how to reproduce it, and which version (the release you downloaded; Help › Report a problem fills it in). We acknowledge reports within three working days, keep you informed while we fix it, and credit you in the release notes unless you prefer otherwise.

Security fixes go into the latest release. archdraw updates itself on macOS and from the AppImage on Linux; deb users are told when a new version is out.

• AI keys are encrypted with the OS keychain (Keychain on macOS, Secret Service or KWallet on Linux) and stored in an owner-only file. On a Linux desktop without a keyring, the app says so: the key then has only basic protection.
• Your code never leaves your machine except to the AI provider you configured, and only what the agent reads for a check or a conversation. The agent's tools are read-only and refuse .env files, keys, credential folders, Terraform state and anything .gitignore excludes.
• The desktop app runs its own server on 127.0.0.1 with a per-launch secret; nothing listens on other interfaces. Diagram SVG is sanitized before display and the page runs under a strict Content Security Policy.
• Git runs with hooks, fsmonitor and ext:: transports disabled.
• No telemetry. archdraw contacts only your AI provider, GitHub (for your repos and for updates) and nothing else.