Apple Intelligence or Open Source?
| Consideration | Apple Intelligence | Open Source |
|---|---|---|
| Requirements | Supported Mac, macOS 26+ | macOS 14+, Apple Silicon recommended |
| Download | No separate model file | A local GGUF file |
| Model choice | Managed by Apple | Curated or user-selected |
| Offline | Yes when available | Yes after download |
curated local models
| Model | Approximate size | Starting guidance |
|---|---|---|
| tabby-2-nano | 0.8 GB | Fastest option for older or lower-memory Macs |
| tabby-2-mini | 1.4 GB | Balanced everyday starting point |
| tabby-2-base | 4.5 GB | Larger local model with higher memory use |
| tabby-2-pro | 5.0 GB | Largest curated model and the heaviest local option |
A larger file is not automatically better autocomplete. If it cannot keep up with normal typing, a smaller model produces the more useful experience even when its individual completions are less ambitious.
bring another GGUF model
- Open Settings → Engine & Model.
- Use the built-in Hugging Face browser or open Cotabby's models folder.
- Download or copy a compatible
.gguffile into that folder. - Choose Refresh, then select the discovered model.
reuse LM Studio models
Enable “Also Use LM Studio Models” in Engine & Model to include GGUF files already stored in the local LM Studio library. Cotabby reads the files locally; it does not require LM Studio to generate each suggestion.
If LM Studio moves or removes a file, refresh Cotabby's model list and select an available model again.
switch by power source
Cotabby can select a different engine or local model on battery and while plugged in. A smaller battery model can reduce memory and energy use while a larger plugged-in model prioritizes completion quality.
Configure both profiles before enabling the switch so moving between power states never lands on a missing model.
offline and network behavior
In the public release documented here, generation uses Apple Intelligence or an in-process local GGUF model. Network access is used for update checks, Hugging Face search, and model downloads—not for the text-generation request itself.
Read the full security and data-flow model or troubleshoot a model that will not load.
related help
Getting Started with Cotabby on Mac
Install, grant permissions, choose an engine, and test your first suggestion.
Customize Cotabby Writing and Context Settings
Length, languages, spelling, multiline output, Fast Mode, and optional context sources.
Troubleshoot Cotabby Autocomplete on Mac
A symptom-first checklist for suggestions, permissions, models, and app compatibility.