Reference
prose — Nemotron 120B
code — DeepSeek v4 Pro
translate — Llama 3.1 8B
pearl image --backend nim — FLUX.1-dev
pearl speak --backend nim — Magpie TTS
/provider nim — switch to NIM
chat — Qwen 2.5 14B GPU
vision — Qwen 2.5 VL 7B GPU
pearl image --backend comfyui — Qwen Image
pearl image --model illustrious-xl — SDXL anime
pearl image --model qwen_image_layered — Layered
/provider ollama — use Pearl Star
Ollama on CPU = ~6 tok/s WARN
pearl status — check placement
/ps/vram/free — unload to free GPU
/model <name> — switch model
/provider nim — switch to NIM
/provider ollama — switch to local
/mode prose — prose contract
/mode code — coding mode
/mode translate — CJK rules
/mode vision — Qwen VL
/models — list available
/status — VRAM + services
/clear — reset history
/help — full command list
Type "use NIM" or "use local" to switch provider naturally.
Image: --backend nim | comfyui
— --model qwen_image | illustrious-xl | layered
— --lora mira_illustrious_lora
TTS: --backend nim | cosyvoice
— --voice en-US | zh-TW | zh-CN | ja-JP | ko-KR