Pulling models¶
heim library pull copies a model from a local source store (the Hugging Face cache,
Ollama, or LM Studio) into the library.
heim library pull <source> <name> # source: huggingface | ollama | lmstudio
heim library pull lmstudio lmstudio-community/gemma-4-12B-it-QAT-GGUF
heim library pull huggingface unsloth/SmolLM2-135M-Instruct-GGUF
heim library pull ollama gemma4:31b
See what's available vs already pulled¶
heim library sources shows models in your app caches with an IN LIBRARY column
(✓ = already pulled) and a summary like 12 models · 3 already in library · 9 to
pull. It's the "what can I pull" view; heim library ls is the "what
do I have" (your library) view.
Move instead of copy¶
--move deletes the local source after a verified, byte-for-byte copy — use
it to relocate models onto the drive and free local disk:
heim library pull lmstudio <name> --move
heim library pull ollama gemma4:31b --move # preserves blobs shared with other models
Source support
--move is implemented for LM Studio and Ollama today. For Ollama it
removes the manifest and only the blobs no other model still references.
Hugging Face pulls download from the hub, so --move there is not applicable.
Per-format destinations¶
LM Studio and HF land in the format-centric layout (gguf/, mlx/, …); Ollama
keeps its native store. Either way the model shows up in heim library ls and is
runnable. See The library for the full layout.