Custom Proxy Support

[dongguowu]

Using a Custom Proxy

This server supports routing requests through any OpenAI-compatible API proxy instead of connecting directly to Google's Gemini API.

To use a custom proxy, set the following environment variables in your .env file:

CUSTOM_PROXY_URL: The base URL of your proxy endpoint (e.g., https://my-proxy.com/v1).
CUSTOM_PROXY_API_KEY: The API key for your proxy service.

When CUSTOM_PROXY_URL is set, the GEMINI_API_KEY is no longer required. The server will automatically:

• Route all completion requests to your proxy's URL.
• Use the CUSTOM_PROXY_API_KEY for authentication.
• Adjust model names by removing the gemini/ prefix to ensure compatibility with standard OpenAI-like proxies.
• Switch to an OpenAI-compatible tokenizer (tiktoken) for the /v1/messages/count_tokens endpoint.
• The server will now send requests to https://api.example.com/v1 instead of Google's servers.

see https://github.com/dongguowu/gemini-for-claude-code/tree/custom-proxy-support

[dongguowu]

Hi, I hope you are doing well. The correct .env for gemini-balance looks like: # GEMINI_API_KEY (Deprecated): A single key or a comma-separated list of keys. # This will be removed in a future version. #GEMINI_API_KEY="" CUSTOM_PROXY_URL="https://your.proxy.url/v1" CUSTOM_PROXY_API_KEY="sk-your-key"

…

On Sun, Aug 17, 2025 at 10:01 AM LavikLee ***@***.***> wrote: *LavikLee* left a comment (coffeegrind123/gemini-for-claude-code#22) <#22 (comment)> I added the two variables CUSTOM_PROXY_URL and CUSTOM_PROXY_API_KEY to my .env file, but it seems that server.py cannot recognize them properly. It keeps showing "GEMINI_API_KEY error" and fails to forward the requests to the address specified by CUSTOM_PROXY_URL. — Reply to this email directly, view it on GitHub <#22 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AJCQOAZVZKSSS43RHGZPLFL3OCDLVAVCNFSM6AAAAACDP2ZDCGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTCOJUGQYDKNRWGY> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[LavikLee]

Thanks a lot. I have already changed server.py to your version, but it still doesn't work. I think the reason is that server.py cannot parse the request into the format accepted by my gemini-balance. For example: https://rlkvpmfancki.ap-southeast-1.clawcloudrun.com/v1beta/models/gemini-2.5-pro:generateContent
It's clear that my gemini-balance only works for the native Gemini API mode and doesn't support the OpenAI compatible mode.
p.s. Ultimately, I identified the issue: the base_url format provided needed to comply with Gemini-Balance's requirements. While the integration works correctly with 'claude code -- gemini for claude code -- gemini balance --gemini ai', the comsumption of Claude Code is extremely high, and Gemini Balance has a very high error rate.