We just added support for the new Gemini Interactions API into Discourse AI and recommend you use it for all Gemini Inference.
The generateContent API remains in Discourse and is now labeled Google Gemini (Legacy API)
All new Gemini LLM providers should use the Google Gemini provider and the: https://generativelanguage.googleapis.com/v1/interactions endpoint.
Caveats
The new interactions API no longer allows setting “no thinking” on 2.5 flash models. This was never advisable for reasoning models anyway.
New Features
The new provider ships with 3 new feature:
- Tool call streaming (which improves web artifact interactivity)
- Code execution (native tool)
- Google Maps (native tool)
In addition to the existing (Search and Fetch tools)
Code execution:
Maps:
