Skip to content
Bedrova bedrova
FeaturesCompatibilityDevelopersCompareDocs Coming soon

Docs

Import local models

Add an MLX folder, GGUF text model, or Core ML perception package you already have — inspection, trust state, repair, and the portable sidecar.

If you already have a model on disk — downloaded elsewhere, copied from another Mac, or on an external drive — you can import the folder directly instead of downloading again.

Importing a folder

Point Bedrova at the model directory or file. It inspects the import the way it reads a Hugging Face repo: weights, tokenizer, config, chat template, runtime format, and capability metadata. A valid MLX model folder has safetensors weights plus tokenizer/config files; a GGUF text import needs runnable llama.cpp metadata; a Core ML perception package needs model provenance and task metadata. If something’s missing or unrecognized, the import preview says what and offers a repair action rather than failing silently.

Trust and quarantine

An imported model starts quarantined until its checks pass, the same as a download:

  • Tokenizer, config, and template are validated before the model can load.
  • Models that would need trust_remote_code are held back — that’s off by default.
  • Once checks pass, the model joins your Local Models with a clear trust state.

The full trust model is in Model trust & verification.

Reusing the portable sidecar

If the folder contains a bedrova.model.json sidecar, the import reuses it — profile, preferences, provenance, and compatibility notes come along, so a model you tuned on another Mac arrives already set up. See Model profiles & presets.

After import

The model appears in your Local Models inventory with its size, trust state, and a profile you can edit. Pin it, set per-model defaults, or load it like any other model. Models on an external drive use security-scoped bookmarks so access survives restarts — see Data & file locations.

See also