Skip to main content

Local Model Setup (Docker-based)

WARNING Local models are currently best for testing and experimentation. Do not expect high performance from local models right now, especially on macOS where GPU acceleration support is limited for this workflow. If you want the best speed and quality, use a remote model provider.

Sulla Desktop supports local models through the Docker-based local model workflow.

When to use local models​

Use local models when you want to:

  • Test local-only workflows
  • Experiment with model behavior
  • Prototype without relying on a remote API

For production-quality agent performance, remote models are still recommended.

How to open local model settings​

Use the app menu path below:

  1. Open the main Electron menu.
  2. Open Language Model Settings.
  3. Go to Models.
  4. Open Local Models.
  5. Choose and configure the local model you want to run.

How to choose a local model​

For better local speed, choose models with:

  • Lower parameter counts
  • Lower context windows

These settings generally reduce resource usage and improve responsiveness on local hardware.

Qwen-family models are often among the better-performing local options.

That said, this is still "best of local" performance, which may not match remote model quality or speed.

Performance expectations​

  • Local model speed depends heavily on your machine resources.
  • Larger models can be noticeably slower.
  • Mac performance may be limited for this local setup.

If agent quality or response speed is not acceptable, switch to a remote model in Language Model Settings.