Clabrate’s AI features — enhancement, transforms, and the assistant — run on a model you choose. You can keep everything on your Mac, or connect a cloud provider for more capable models. You decide, per profile.
Local vs. cloud
- Local (on-device) — the model runs entirely on your Mac. Nothing leaves the device. This is the default and the private path.
- Cloud — the model runs on a provider’s servers. More capable for some tasks, but the text for each request is sent to that provider.
Your choice: Stay fully on-device and Clabrate never sends your text anywhere. Cloud is opt-in. See Privacy & your data.
The providers
| Provider | Where it runs | Notes |
|---|---|---|
| Ollama | Local — http://localhost:11434 | Run open models on your own machine |
| LM Studio | Local — http://localhost:1234 | Another local model runner |
| Gemini | Cloud (Google) | API key from aistudio.google.com/apikey, stored in your Keychain |
Set these up in Settings → Brain & models.
Screenshot pending: providers-settings.png
Model profiles
A profile bundles a provider and model so you can switch with one click. Two are built in, and you can create your own:
- Llama 3.2 1B — on-device, the default. Fast, private, no internet needed.
- Clabrate Smart — Gemma 4 E4B, an optional ~6 GB on-device download. More capable, and image-capable (useful for screenshot attachments in the assistant).
- Your own profiles — point at any provider/model you’ve configured (e.g. a local Ollama model, or Gemini).
Note: The bigger the model, the more capable — and the slower / heavier. The 1B default is tuned to be quick for everyday enhancement; reach for Clabrate Smart or a cloud model when a task needs more.
Common questions
Do I need an API key? Only for cloud providers like Gemini. Local providers and the built-in on-device models need none.
Where’s my key stored? In the macOS Keychain — never in a log or Clabrate’s database.
Which should I pick? Start on the on-device Llama 3.2 1B default. Add Clabrate Smart for tougher rewrites or image context; add a cloud provider only if you want it.
Troubleshooting
- A local provider won’t connect. Make sure Ollama / LM Studio is running and listening on the expected port.
- Gemini is unavailable. Check your network and that the API key is entered — see Troubleshooting.