Skip to Content
Model GuideIntegrationsGemini CLI

Gemini CLI

Gemini CLI  can override its Gemini API endpoint when using API-key authentication. Use Fenno’s native Gemini host, not its OpenAI base URL.

1. Prepare the CLI and credentials

Install Gemini CLI using its official installation instructions , then check:

gemini --version

Create a Fenno key, select a model supporting Gemini content generation, and set FENNO_API_KEY and FENNO_MODEL locally. Verify the chosen model with the Gemini API reference first.

2. Configure API-key authentication

export GEMINI_API_KEY="$FENNO_API_KEY"
export GOOGLE_GEMINI_BASE_URL="https://api.fenno.ai"
gemini --model "$FENNO_MODEL"

If prompted, choose Use Gemini API key, not Google account login or Vertex AI. The base-URL override applies to this API-key path. Do not append /v1, /v1beta, or another provider’s /gemini prefix.

If you previously used Vertex AI or another endpoint, review conflicting environment variables and saved authentication. Avoid replacing an entire settings.json just to change one setting. Keep keys out of committed .env files.

3. Verify before granting tools

Start with a minimal text request:

gemini --model "$FENNO_MODEL" --prompt "Reply with OK only."

Check the request time, model and usage in the Fenno account records available to you. A model saying it uses Fenno is not evidence of the actual provider.

Then test streaming and tools separately in a disposable workspace, keeping normal permission prompts. Do not enable unrestricted execution merely to test connectivity.

Compatibility boundaries

  • Gemini CLI may use additional API methods as it evolves. A successful generateContent call is not proof every CLI feature works through a gateway.
  • A feature requiring an unprovided method, hosted search or model-specific tool cannot be enabled just by changing the base URL.
  • For 401, verify the selected authentication path and environment inherited by the process.
  • For 404, inspect the final endpoint and model ID; for tool failures, check model support before changing unrelated settings.

See the official configuration reference , Gemini endpoints, and generic troubleshooting.

CursorZed