You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
- Use environment variables for security (recommended)
177
+
- Override per-environment using config files
178
+
- Mix different approaches for different providers
179
+
180
+
#### Provider-Specific Notes
181
+
182
+
-**PPInfra**: ✅ No API key required - uses public API
183
+
-**OpenRouter**: ✅ No API key required - uses public model listing API
184
+
-**Gemini**: ⚠️ Optional API key - uses hybrid web scraping + API approach
185
+
186
+
### Gemini Provider Details
187
+
188
+
The Gemini provider implements a unique **hybrid approach**:
189
+
190
+
**How It Works:**
191
+
1.**API Call**: Fetches model list from Gemini API (model names only)
192
+
2.**Web Scraping**: Scrapes Google's documentation for detailed capabilities
193
+
3.**Data Merging**: Combines API data with scraped metadata
194
+
195
+
**Behavior by API Key Status:**
196
+
-**With API Key**: Complete model list from API + enriched capabilities from scraping
197
+
-**Without API Key**: Model list and capabilities from web scraping + fallback known models
198
+
199
+
**Why Hybrid?** The official Gemini API only provides model names, so web scraping is always required to get comprehensive capability information (vision, function calling, reasoning, context lengths, etc.).
200
+
149
201
## 🤖 GitHub Actions Automation
150
202
151
203
The project includes GitHub Actions workflow with multiple trigger methods:
@@ -225,8 +277,8 @@ For detailed development guide, see [Architecture Documentation](docs/architectu
225
277
226
278
- ✅ **PPInfra** - 38 models with reasoning, function calling, and vision capability detection
227
279
- ✅ **OpenRouter** - 600+ models with comprehensive capability detection and metadata
228
-
-🚧**OpenAI** - Planned
229
-
- 🚧 **Google Gemini** - Planned
280
+
-✅**Google Gemini** - Gemini models with hybrid API + web scraping approach for complete metadata
0 commit comments