Google Gemini

Gemini models for text, the Live API for voice (with a live camera view), and Gemini's text to speech.

Key: create one in Google AI Studio (aistudio.google.com/apikey).

What it's used for

Use	Supported
Text replies (the brain)	Yes
Images and camera frames	Yes
Tools	Yes
Realtime voice	Yes: the Live API (voice.mode: realtime with a Gemini connection)
Live camera video	Yes: with vision.webcam.mode: continuous, Gemini Live gets a frame about once a second while the picture changes
Speech to text (cascade)	No: use OpenAI or ElevenLabs
Text to speech (cascade)	Yes: gemini-3.8-flash-lite-tts when no model is chosen

Models

The server reads your key's model list. New bots start with the newest gemini-N-flash model for text, and a Live model for voice.

• On Gemini 3 and later, brain.effort sets the thinking level: medium and high mean high, minimal and low mean low. Older models don't get a thinking level.
• brain.temperature is passed through when set.

Voices

Gemini's fixed set of named voices (Puck, Charon, Kore, Fenrir, Aoede, Leda, Orus, Zephyr and more), with a description of each in the admin panel's list. In the cascade, voice.cascade.tts.style is passed to the voice as a style prompt; speed isn't supported by Gemini.