Please do not open a public issue for security problems. Report them privately to the repository owner through GitHub's private vulnerability reporting feature.

• Never commit .env files, API keys, logs, or model responses.
• The repository previously tracked a beta .env and a server log. Both have been removed from the tracked tree. Treat any key that appeared in git history as compromised: revoke it at the provider first, then purge history with scripts/purge_secrets.sh, force-push, and audit provider usage.
• The server logs prompt lengths instead of prompt bodies at INFO level. Set LOG_LEVEL=DEBUG only for local debugging.
• Bring-your-own-key settings live in %APPDATA%\Vulcan\settings.json. ServerToken and LlmApiKey are encrypted with Windows DPAPI under the current Windows user, so the file must not be copied between machines or accounts and must never be committed. The key is forwarded to the configured Vulcan server as X-LLM-API-Key for a single request and is not logged by the server.
• Set VULCAN_ALLOW_CUSTOM_LLM_BASE_URL=false on a shared server to prevent clients from redirecting prompts and keys to an arbitrary provider endpoint.

• Set VULCAN_API_TOKEN and require Authorization: Bearer <token> on the server.
• Keep HOST=127.0.0.1 unless a trusted reverse proxy terminates TLS in front of the service.
• Do not enable Flask debug mode in production. The Docker image runs gunicorn with FLASK_ENV=production.
• Per-IP rate limiting is in-memory and per worker. Use a shared store such as Redis when running multiple workers or replicas.