hermes - 💡(How to fix) Fix [Bug]: Vision module routes Gemini API calls through native adapter even when OpenAI-compatible endpoint is configured

Description

When configuring the auxiliary vision module to use Gemini models via the OpenAI-compatible endpoint (/v1beta/openai), Hermes's Gemini native adapter still intercepts the request and converts it to the native Gemini REST API format, causing a GenerateContentRequest.model: unexpected model name format error.

Steps to Reproduce

1. Set environment variable GOOGLE_API_KEY with a valid Gemini API key
2. Configure vision module:

auxiliary:
  vision:
    provider: gemini
    model: gemini-2.5-flash
    base_url: https://generativelanguage.googleapis.com/v1beta/openai
    api_key: <key>

1. Call vision_analyze with any image → Error 400

Root Cause

is_native_gemini_base_url() in agent/gemini_native_adapter.py should return False when base_url ends with /openai, but in the resolve_vision_provider_client() flow, the native adapter is triggered regardless.

Even with explicit base_url: .../openai, requests are converted to Gemini native :generateContent format instead of OpenAI-compatible /chat/completions.

Verified Working Workaround

Direct API calls work fine:

url = "https://generativelanguage.googleapis.com/v1beta/openai/chat/completions"
headers = {"Authorization": "Bearer <KEY>"}
body = {"model": "gemini-2.5-flash", "messages": [...]}
# → 200 OK

Environment

• Hermes Agent v0.14.0
• Python 3.10
• Ubuntu 22.04