Persist the user message before streaming so the model receives the current question (previously only the system prompt was sent on new chats, causing canned greetings). Also fixes user message being lost when the AI call errors.
- Cookie-based auth via backend proxy (httpOnly JWTs) - Supabase Postgres persistence for sessions/messages/profiles + RLS - Fix cross-user session leak (token cache keyed by full token, not 50-char prefix) - Fix missing table grants (42501) via migration; auto-provision profiles on user creation - Chat validation, ownership checks, rate limiting, /api/chat/sessions route ordering - Frontend auth-state reset + credentials include - OpenRouter AI provider (OpenAI-compatible base URL, reasoning disabled) - Tests: chatValidation, appState
hahachait