Local

In-process local VLM (llama.cpp + libmtmd): a GGUF model loaded directly into the process, no external server. Configured via .