Google Gemini
Connect a Google AI API key and use the Gemini model currently available in Mobius.
Use the Google Gemini integration when you want managed Gemini model calls to use your organization's Google AI API key. The key takes precedence over a platform Gemini credential or a platform Vertex route.
This integration stores a model key. It does not itself add an assistant action
or source event. Built-in gemini.* media actions are a separate capability;
check the action catalog before trying to add or use them.
Save a Google AI key
Open Build > Integrations > Google Gemini, select Connect Google Gemini, enter the Google AI API key, and select Save credentials.
The API equivalent is
POST /v1/integrations/providers/gemini/connect with api_key. Use gemini as
the provider ID.
Mobius currently offers two Gemini models for a new assistant or conversation override:
| Model | ID | Guidance shown in Mobius |
|---|---|---|
| Gemini 3.8 Flash | gemini-3.8-flash | Fast multimodal work |
| Gemini 4 Argon | gemini-4-argon | Flagship reasoning |
Older Gemini entries remain known so existing assignments can be interpreted, but the current picker does not offer them for a new choice.
Verify the route
After connecting, confirm that Google Gemini appears in the assistant's
Model menu and that the catalog source is byok. BYOK model calls use the
connected account's quota. A platform source means Mobius supplies the route,
and the turn must pass plan and credit checks when it runs.
Create a new conversation for the test. A model selected there becomes a saved conversation override. If you leave the choice unset, the conversation follows the assistant's current saved model. Worker-routed assistants do not accept a managed model override.
If the model is visible but a turn fails, read the turn error. Catalog availability records configured credentials; it does not pre-approve platform funding for the later call.
Integrations explains organization connections. Read Actions before granting a separately registered Gemini media action.