Goal

Downloading

• Users can cortex pull from Cortex curated model repos
• Users can cortex pull from HF URL
• Users see rich logs informing progress and success/failure

Importing

• Users can "import" a local model for Cortex

Model YAML

• The correct default settings are inferred for the model
• Related: bug: Dropping GGUF metadata when creating model.yaml for Model Downloaded via URL jan#3558

Model Alias/Branch

• Do we still support alias?

Questions

1. Should we cache downloads so users can resume them?
2. Should we auto scan popular folders for existing models?
3. Should users be warned about downloading incompatible models?
4. HF download often times out, leading to corrupted state.

Related

• feat: Cache model and engine downloads so users can restart download #777
• https://github.com/janhq/jan/issues/3521
• bug: cortex pull from hf times out #1017
• bug: Download models/engine failed cause api server stop #1010
• feat: Add pause functionality to model downloader jan#3519
• feat: Cortex model importing #1058
• epic: cortex models pull <ID_or_URL> #1077
• idea: cortex runs with model alias #1008
• feat: Cortex pull from HF can select quantization #1006
• feat: cortex pull supports full HF URL download #1004