Google Gemini
A Google AI Studio key, or your Google account through Antigravity.
In short
The Gemini engine uses Google's native API with a key from Google AI Studio. It handles long context and images, so it can read tables and figures for Vault Vision. You can also reach Google's models without a key through Antigravity (agy).
1Get a key
Open Settings → AI engines → Google Gemini → Get a key to reach Google AI Studio, then Add Gemini and paste the key.
2Pick a model
Free-tier keys have a daily request limit per model. If a model runs out, the app says which quota was hit and offers a switch.
For technical users
Wire: POST /v1beta/models/{model}:streamGenerateContent?alt=sse with the key in the x-goog-api-key header (never ?key=, because URLs reach logs). The native API is used rather than Google's OpenAI-compatible endpoint because it carries the reasons the UI must show: promptFeedback.blockReason, finishReason categories, RetryInfo for a per-minute 429, QuotaFailure for a per-day quota, and API_KEY_INVALID. Errors add SAFETY_BLOCKED and QUOTA_EXHAUSTED to the shared set. Images are sent as inlineData parts. The second row, Use your Google account via Antigravity (agy), points at the agy CLI engine and needs no key.