Module config 

Persisted model selection.

Which GGUF the daemon should load is stored in a one-line file in the data dir, so the choice survives a daemon restart without depending on a shell env var (KINTSUGI_MODEL_FILE). kintsugi model use writes it; the daemon’s LlamaScorer::autoload reads it after the env override. This is the bring-your-own-weights path: any GGUF works, so a model can be swapped at any time without updating Kintsugi itself.

Always compiled (not behind llama) so the CLI can manage the selection even when the installed daemon has no inference engine.

Functions

clear_configured_model

Forget the persisted selection. A no-op if none is set.

configured_model

The persisted model path, if one is recorded. Returns the path even when the file no longer exists on disk, so callers can report a stale selection rather than silently ignore it.

model_config_path

The file recording the configured model path (one line: an absolute path).

set_configured_model

Persist path as the configured model, creating the data dir if needed.