The supported alpha deployment is one user on a trusted computer. The default server binds to 127.0.0.1. Local browser requests may bypass login only when the network peer and browser Origin are trusted loopback values.
LAN mode is an advanced deployment, not a multi-user security boundary. All authenticated users would access the same meetings and people.
- No wildcard CORS in the default configuration
- HTTP and WebSocket Origin validation for local bypass
- JWT authentication outside trusted local bypass
- First-login password change for the default account
- Rate limits and basic browser security headers
- Public LLM endpoints require explicit opt-in
HOST=0.0.0.0
DEPLOYMENT_MODE=lan
LOCAL_AUTH_DISABLED=false
JWT_SECRET=<at-least-32-random-bytes>
ALLOWED_ORIGINS=https://matrix.example.internalPut the service behind an HTTPS reverse proxy and firewall. Do not expose port 8000 directly to the internet. Keep WORKERS=1; inference engines and the in-process job runner are designed for one process.
Audio, transcripts, embeddings, SQLite files, .env, JWT secrets, and LLM keys must never be committed. API keys saved in Settings are stored in SQLite without application-level encryption. Prefer a local LLM or an environment variable when stronger secret isolation is required.
Model loaders download third-party artifacts and some ModelScope packages may execute trusted upstream model code. Review model sources and licenses, use isolated environments, and avoid running unreviewed cache contents with elevated privileges.
Do not open a public issue containing recordings, transcripts, credentials, or exploit details. Report vulnerabilities privately through GitHub Security Advisories. Repository maintainers must enable private vulnerability reporting before release.