Load a model to serve it on SecLLM's OpenAI endpoint. Several models run at once, packed onto your GPUs by available VRAM — Load a new one and it coexists with the others (set
SECLLM_MAX_LOADED=1 if you instead want loading a model to switch by evicting the current one). When the GPUs are full a Load is refused rather than crowding a card. Point SecRouter at http://<host>:11400/v1. The context field overrides that model's default context length (tokens) for this load only — blank uses the catalog default. Download pre-fetches a model's weights without loading/serving it — useful for warming several models ahead of time; Load downloads automatically too if you skip this.Models
sign in with SecSSO, or connect with an admin token, to manage models