Subjects, notes, LaTeX & an AI study partner.
Keys are stored in your own Firestore document and used only from this browser. On each request the app tries Gemini first, then Groq, then OpenRouter — whichever have keys. Within a provider it works through your models list top to bottom, and for each model rotates through your keys — one key or model failing (quota, overload, retirement) just moves to the next.
Real numbers from a live account, requests/day — each model is a separate quota bucket even on the same key, so listing several stacks their limits instead of sharing one: gemini-3.5-flash-lite / gemini-3.1-flash-lite: 500/day · gemma-4-31b-it / gemma-4-26b-a4b-it: 14,400/day (same Gemini key & endpoint, just open-weight models) · flagship Flash (3.5/3.6/3.7/3.8 incl. gemini-flash-latest, and 2.5): only 20/day each, but still free extra requests once the rest above are exhausted for the day. gemini-2.5-pro has zero free quota — don't bother.
ChatGPT isn't offered here — OpenAI's chat API blocks direct browser calls (no CORS on its response), so it needs a backend server to proxy it, which this app doesn't have.