Gemini CLI：版本与协议
Google sign-in and a NekoCloud API key are different authentication paths. The official API-key tutorial targets Google AI Studio; replacing its key alone does not prove third-party connectivity.
CC Switch and ccman expose Gemini CLI configuration using variables such as GOOGLE_GEMINI_BASE_URL and GEMINI_API_KEY, but endpoint handling varies by CLI version. Inspect your version and generated settings. Use API-key mode, not Google OAuth or Vertex. Match native generateContent protocol, model permissions and /v1beta handling; verify whether the SDK appends the API version.
No unverified fixed launch command is provided here. To use a Gemini model without requiring Gemini CLI, choose a client with a configurable matching protocol. ACP authentication and environment inheritance require separate verification.
Verification and safety
Back up existing settings and merge only relevant fields; preserve other providers, plugins, MCP and permissions. Protect keys and backups, never commit them. Variables must reach the actual agent process; Remote SSH, containers and desktop processes have separate environments.
Relaunch, select the NekoCloud model, send a short message and match time, model and charge in Usage. Then read a non-sensitive file in an isolated project and verify the tool call, returned result and final answer. Tests can incur charges. Keep approval controls enabled.
This is adapted from official provider configuration guidance, not proof that every model or the complete NekoCloud tool chain has been tested. Check Model Plaza, key groups and troubleshooting.