Gemini CLI

Use Gemini CLI with gateway API-key authentication and the native v1beta API.

Install and prepare

  1. Install Node.js 20 or later, then the official Gemini CLI npm package below. Check gemini --version.
  2. In routerra.ai API Keys, create a key with access to a native Gemini model from Models. This connection uses Gemini generateContent/streamGenerateContent, not OpenAI Chat Completions.
npm install -g @google/gemini-cli
gemini --version

Configure native Gemini routing

  1. Set the three variables below in the terminal launching Gemini CLI. The SDK appends /v1beta/models/..., so GOOGLE_GEMINI_BASE_URL must be https://routerra.ai without /v1 or /v1beta. Replace the key and model placeholders.
  2. When prompted, choose Use Gemini API key; newer builds may detect gateway authentication automatically. An existing Google-login or Vertex session can bypass this route: use /auth to change it. For persistence, save the variables as NAME=value lines in ~/.gemini/.env, then restart Gemini CLI. Do not commit that file.
export GEMINI_API_KEY="YOUR_API_KEY"
export GOOGLE_GEMINI_BASE_URL="https://routerra.ai"
export GOOGLE_GENAI_API_VERSION="v1beta"
gemini --model YOUR_GEMINI_MODEL_ID

For Windows PowerShell:

$env:GEMINI_API_KEY = "YOUR_API_KEY"
$env:GOOGLE_GEMINI_BASE_URL = "https://routerra.ai"
$env:GOOGLE_GENAI_API_VERSION = "v1beta"
gemini --model YOUR_GEMINI_MODEL_ID

Check the connection

Test a short prompt with the explicit --model ID. A Google sign-in page indicates the wrong authentication mode. A request to /v1/v1beta or /v1beta/v1beta means the base URL contains an extra path. If the selected model is rejected, check its native Gemini availability instead of substituting an OpenAI model ID. Remove conflicting GOOGLE_GENAI_USE_VERTEXAI or GOOGLE_GENAI_USE_GCA settings, restart, and confirm usage in routerra.ai.