Gemini CLI
The gemini-3-pro-preview model is available in the NexoAI catalog, but the stock Gemini CLI speaks Google's Gemini-native API. NexoAI's documented client contracts are the OpenAI-compatible and Anthropic-compatible shapes, so a base-URL override alone does not translate Gemini-native requests into either supported shape.
Compatibilité, sans deviner
Do not point GOOGLE_GEMINI_BASE_URL at NexoAI or put a NexoAI key in GEMINI_API_KEY unless the NexoAI dashboard explicitly announces Gemini-native API support. Those variables change the destination, not the request schema.
Model availability is not protocol compatibility
Seeing gemini-3-pro-preview in the NexoAI model catalog means you can call that model through a supported NexoAI API shape. It does not make every Google-native client OpenAI- or Anthropic-compatible.
Use the Gemini model today
Configure an OpenAI-compatible tool with the NexoAI base and key:
bash
export OPENAI_BASE_URL="https://api.nexoai.ma"
export OPENAI_API_KEY="sk-your-key"Then select gemini-3-pro-preview as the model in that tool. The curl, Python, and Node.js examples all work with that model by changing the model field.
For an agentic terminal workflow, use Codex with the nexoai provider and set:
toml
model = "gemini-3-pro-preview"When direct setup becomes available
Direct Gemini CLI support requires a documented Gemini-native endpoint at the gateway, not just a catalog model. When NexoAI exposes that contract, this page should specify the exact CLI version, authentication mode, base variable, and a tested health request. Until then, the OpenAI-compatible route is the supported path and avoids silently sending a NexoAI key to the wrong service.
Never test a key against an unknown host
Before placing a NexoAI key in any third-party variable, confirm that the configured base URL is https://api.nexoai.ma and that the client uses a supported API shape.
Continue with generic OpenAI-compatible tools →