Uploaded October 2025 | Updated September 2026, 6 hours ago
- Classic SSE reverse proxy
- Made while waiting for llama.cpp to handle model loading natively I couldn't fix llama-swap's reliability issues
- Uses the same base directives as config.yaml and runs with the same command line structure
- Sends extra chunks inside reasoning_content so the user instantly sees that the model weights are loading instead of staring at a blank screen. And since this doesn’t break OpenAI-compatible semantics or pollute the context, it just works smoothly
- No need for a separate web interface anymore, all essential info is already displayed
- Classic SSE reverse proxy
- Made while waiting for llama.cpp to handle model loading natively I couldn't fix llama-swap's reliability issues
- Uses the same base directives as config.yaml and runs with the same command line structure
- Sends extra chunks inside reasoning_content so the user instantly sees that the model weights are loading instead of staring at a blank screen. And since this doesn’t break OpenAI-compatible semantics or pollute the context, it just works smoothly
- No need for a separate web interface anymore, all essential info is already displayed
