qwen2.5vl:3b
For speedThe fastest first choice for everyday screenshots and images. Use it when you want local processing to feel responsive on most Macs.
Most Apple Silicon MacsFastest
ollama pull qwen2.5vl:3bOllama Setup Guide
Use Offline AI mode with private local models via Ollama. Your files are processed by a model on your device instead of a cloud AI provider.
What Offline AI Mode Means
When Offline AI mode is enabled in Zush, supported file analysis runs through your local Ollama server. Zush does not send analysis content to Zush cloud or third-party AI providers in this mode. You still control which model is installed, where Ollama stores it, and when Ollama is running.
Setup Steps
1
Install Ollama
Download Ollama for macOS from the official website, install it, and open the app once so the local server can start.
Open Ollama download page2
Download a vision model
Zush works best with a vision-capable model because many files are images, screenshots, PDFs, or visual previews. Start with:
ollama pull qwen2.5vl:3b3
Check that Ollama is running
Ollama usually runs in the background after you open it. If Zush cannot connect, start it from Terminal:
ollama serve4
Enable Offline AI mode in Zush
Open Zush, go to AI Setup, turn on Offline AI mode, refresh the model list, select your model, and run Test.
Recommended Models
Pick a model based on the job: qwen2.5vl:3b for speed, gemma3:4b for balance, or granite3.2-vision:2b for documents.
gemma3:4b
For balanceA good default when you want better descriptions without jumping to a large model. Start here if speed and quality both matter.
8 GB+ memory recommendedBalanced
ollama pull gemma3:4bgranite3.2-vision:2b
For documentsA compact vision model that is useful for document previews, scans, and structured visual content.
Most Apple Silicon MacsFast
ollama pull granite3.2-vision:2bOr choose another vision-capable model from the Ollama model catalog .
Troubleshooting
Zush does not see any models
Run ollama list in Terminal. If the list is empty, pull a model first, then click refresh in Zush.
Connection test fails
Make sure Ollama is running and the host is set to http://127.0.0.1:11434 in Zush connection settings.
Processing is too slow
Use a smaller model like qwen2.5vl:3b, close memory-heavy apps, or switch back to Cloud when you need faster batch processing.
Offline AI mode is separate from Cloud and BYOK. Cloud uses Zush credits by default, BYOK uses your provider key for cloud renames, and Offline AI mode uses Ollama on your device.