Shift note: keep Gemini CLI on the native relay route

The handoff risk is straightforward: changing Gemini CLI’s destination does not automatically change its authentication mode or request format. A relay can accept your key and still reject the request because the client is speaking the wrong protocol. Treat endpoint selection, credential selection, and model access as three separate checks.

For Relay365, the published API base is https://api.api365relay.com. OpenAI clients add /v1. That suffix is not a universal instruction for every tool, and copying it into Gemini CLI is not a substitute for configuring native Gemini routing.

Native Gemini routing keeps the client’s contract intact

A native Gemini relay accepts the Gemini API request structure rather than requiring the client to translate everything into OpenAI-compatible calls. This matters because Gemini CLI does more than send a plain chat prompt: its workflow can involve streaming, tool calls, and structured responses.

The competing approach is an OpenAI-compatible endpoint. That is useful for software built around the OpenAI SDK, but Gemini CLI is not automatically an OpenAI client just because both services expose HTTPS APIs. Pointing it at an OpenAI-style /v1 endpoint does not teach it a different protocol.

The native route wins for an unchanged Gemini CLI. An OpenAI-compatible route wins when your application already uses that interface. For Relay365, confirm the documented native Gemini endpoint and supported CLI features before treating the general API base as a complete integration recipe.

Relay API keys and Google sign-in serve different destinations

Google sign-in is appropriate for the Google-backed authentication flow it was designed to serve. A relay-issued API key is a separate credential, issued and validated by a separate service. Keeping a cached Google login while merely replacing the base URL can leave the CLI using an authentication path the relay does not support.

For a relay deployment, choose Gemini CLI’s Gemini API-key authentication mode and supply the relay key through GEMINI_API_KEY, provided that this is the relay’s documented integration method. Do not assume that an environment variable overrides an authentication choice already saved by the CLI.

API-key mode wins for a documented key-based relay integration. Google sign-in wins when you intend to remain on the corresponding Google service. Neither credential is a drop-in replacement for the other.

A clean shell beats inherited authentication state

Use a fresh terminal session for the first connection. In Gemini CLI versions that support GOOGLE_GEMINI_BASE_URL, set that variable to the relay’s documented native base. Set GEMINI_API_KEY to your relay credential, then explicitly select the API-key authentication option in the CLI.

• Check your installed Gemini CLI version and its documentation before relying on a custom-base environment variable.
• Remove conflicting Google API-key or Vertex AI configuration from the test session without deleting credentials needed by unrelated applications.
• Keep the relay key out of shell history, shared settings files, screenshots, and diagnostic logs.
• Confirm whether the relay expects a root base URL or a versioned path, because duplicating a path segment can break otherwise valid requests.

This isolated setup is easier to diagnose than a long-lived shell containing several providers’ credentials. It also gives you a cleaner rollback: restore the previous environment and authentication selection.

A successful prompt does not clear tool execution

Start with a short request using an enabled model, such as Gemini 2.5 Flash, with the exact identifier supplied by the relay. Then test streaming and a harmless tool-assisted task separately.

An authentication error points toward credentials or auth mode; a missing-route response suggests endpoint construction; a model rejection suggests naming or account access. These are starting hypotheses, not guaranteed diagnoses. Inspect sanitized error details before changing several settings at once.

Relay365 fits consolidation, with native support as the gate

Relay365 is a pay-as-you-go relay desk spanning multiple model families. It fits engineers who want consolidated purchasing across tools such as Gemini CLI, Claude Code, Codex, and OpenAI SDK applications, while still configuring each client’s protocol correctly.

Create a Relay365 account if that consolidation matches your workload, then verify native Gemini instructions and model access before moving an established workflow.

The final verdict is to preserve Gemini’s native protocol, deliberately select relay-key authentication, and validate features beyond a basic prompt. Keep Google sign-in for Google-backed use, and reserve the OpenAI-compatible /v1 route for clients that actually speak it.