Point any OpenAI client at the base URL above. With an API key set, clients must send Authorization: Bearer <key>.

The leading binary name is ignored — the bundled gguf-server engine always runs. Arguments are passed verbatim, including --host/--port.

The model weights served over the OpenAI-compatible API.

Multimodal projector (mmproj) — enables vision input for models that ship one.

Use 0.0.0.0 to accept connections from other devices. Default 8888 — vLLM uses 8000, llama.cpp 8080, LM Studio 1234, Ollama 11434.

The alias is the name reported by /v1/models. Blank uses the file name.

GPU index used for scratch buffers and the final layers.

q8_0 saves VRAM with minimal quality loss.

A preset captures the model, chat template and every server setting.

No saved presets yet.