Manage models on your Mac
Download, verify and maintain the model files used for local inference.
Your space → Models & voice → Local Models
Before you begin
A compatible Mellow build with the relevant capability configured.
Choose a manageable model
Start with a model appropriate for your Mac’s memory. Read the bundle requirements and allow the download to finish.
Select it in chat
Choose the model from the conversation picker. A model being installed is not the same as it being loaded and ready for a request.
Maintain files deliberately
Use the detail view to inspect updates or repair damaged files. Read the operation before confirming it; metadata checks and file verification are different.
Controls & reference
| Control or concept | How to use it |
|---|---|
| Catalog | Find a compatible model and review its size before downloading. |
| On Device | Inspect installed models and their available maintenance actions. |
| Repair | Check files and restore missing or changed content; this can replace deliberate local edits. |
| Check for Model Updates | Compare published revision information without automatically replacing weights. |
| Model residency | Controls when models remain in memory, separately from whether they are installed. |
Try it with a small task
Install a small compatible model, ask for a one-sentence response, then send a follow-up that refers to the first answer.
If it does not work
- Do not infer model compatibility from the filename alone.
- A matching published revision does not establish that local files are unmodified.
- Local inference can still invoke network tools if you enabled them on the agent.
Good to know
- Do not infer model compatibility from the filename alone.
- A matching published revision does not establish that local files are unmodified.
- Local inference can still invoke network tools if you enabled them on the agent.