Please do not open a public issue for a security problem. A public report tells everyone about the flaw before there is a fix available.

Report privately instead, using either:

1. GitHub private vulnerability reporting — go to the repository's Security tab and choose Report a vulnerability. This opens a private advisory visible only to the maintainers. (Repository owners: enable this under Settings → Code security and analysis → Private vulnerability reporting.)
2. Email the maintainers at the address listed on the repository's GitHub profile, with SECURITY in the subject line.

Please include:

• what the issue is and which component is affected;
• the steps to reproduce it, or a proof of concept;
• the impact you believe it has;
• the version or commit you tested.

What to expect: an acknowledgement within 5 working days, an assessment within 10 working days, and credit in the fix notes unless you would rather stay anonymous. This is a volunteer-maintained project, so there is no paid bounty.

Only the latest commit on main receives security fixes.

The defaults in this repository are chosen so a fresh clone runs locally with no configuration. Those same defaults are not safe on a public network. Work through this list before exposing an instance to the internet.

• DEBUG=false. With debug on, Django serves full tracebacks — including settings and local variables — to anyone who triggers an error.
• DJANGO_SECRET_KEY is set to a fresh random value and is not shared between environments. Generate one with: python -c "from django.core.management.utils import get_random_secret_key as k; print(k())". With DEBUG=false the app refuses to start without it; it will never fall back to the development key.
• ALLOWED_HOSTS lists real hostnames. * is rejected outright when DEBUG=false.
• CORS_ALLOWED_ORIGINS lists only your own frontend origins. The API sends credentials, so a wildcard would let any website call it on a visitor's behalf. CORS_ALLOW_ALL_ORIGINS is never enabled by this code.
• RECAPTCHA_SECRET_KEY is set. Required when DEBUG=false. Without it the contact form is an open, unauthenticated write endpoint.
• OPENAI_API_KEY is set, scoped to this application, and has a spend limit. Every recording that reaches the API costs money.

• Terminate TLS in front of the app. With DEBUG=false the app sets HSTS, SESSION_COOKIE_SECURE, CSRF_COOKIE_SECURE, X-Frame-Options: DENY and content-type nosniff automatically.
• If a proxy already redirects HTTP to HTTPS, set SECURE_SSL_REDIRECT=false to avoid a redirect loop.
• Run python manage.py check --deploy and resolve every warning.

• Use the provided Dockerfiles, or an equivalent WSGI server. Never deploy manage.py runserver or the Vite dev server. Both images run as a non-root user.
• Do not serve MEDIA_ROOT from Django in production. The URL pattern for it is registered only when DEBUG is on.
• Restrict /admin/ by network, or disable it if unused. There is no rate limiting on the admin login.
• Tune API_ANON_THROTTLE_RATE (default 30/minute). The audio endpoint is unauthenticated and each call spends money at OpenAI.
• Cap upload size at the proxy too. AUDIO_MAX_UPLOAD_BYTES is enforced by the app, but a proxy limit stops large bodies earlier.

Deploying this application means processing other people's personal data. Know what it collects before you go live.

Notes on the design:

• The session id is not an account. It is a random UUID minted in the browser's localStorage and rotated every 24 hours. It groups one visitor's recordings; it is not tied to a name, an email, or an IP address.
• The server does not log IP addresses itself. Your reverse proxy almost certainly does. That log is personal data too — set a retention period on it.
• Recordings are sent to OpenAI for transcription and text generation, and the generated text is sent to Google Translate's TTS endpoint by gTTS. Both are third-party processors. Say so in your privacy notice.
• The contact form transmits a token to Google for reCAPTCHA verification.

Session data expires on its own, and the sweep runs on every request. To purge sooner, or to run it from a scheduled job:

python manage.py purge_sessions          # delete anything past retention
python manage.py purge_sessions --all    # delete every stored session

For the strongest posture, set AUDIO_PERSIST_RECORDINGS=false. Recordings are then processed and deleted immediately, leaving only the rolling text history.

Contact form submissions have no automatic expiry. If you deploy this, decide on a retention period and delete rows on that schedule — via the Django admin, or a scheduled AppFormSubmission.objects.filter(submitted_at__lt=...).delete().

These are design constraints of the project, not undisclosed vulnerabilities:

• Both API endpoints are unauthenticated. Protection is rate limiting and reCAPTCHA, not identity. Anyone who can reach the audio endpoint can spend your OpenAI quota.
• The session id is client-supplied. It is validated as a UUID, so it cannot be used to escape MEDIA_ROOT, but a caller who guesses or replays another session's UUID would read that session's transcript history.
• VITE_* variables ship inside the JavaScript bundle and are readable by anyone. This is correct for the reCAPTCHA site key; never put a secret there.