Set Up Ollama for Local AI
Use Ollama when you want supported file analysis to run on your computer. Zush guides you through installing Ollama, starting it, and downloading your first model from Settings → AI Modes → Ollama. You do not need Terminal for the default setup.
What you need
Section titled “What you need”- Zush for Mac or Windows, on Free or PRO. Free includes 50 renames shared across all AI modes; PRO removes the rename limit.
- Internet access to install Ollama and download a model.
- Enough disk space and memory for the model. Zush shows the download size when it is available.
Ollama and a model are installed during the steps below. If you already have them, Zush lets you use your existing setup.
Set up Ollama for Zush
Section titled “Set up Ollama for Zush”- Open Zush Settings → AI Modes and choose Ollama.
- If Zush shows Download Ollama, click it to open the official download page. Install the Ollama app, then return to Zush.
- If Zush shows Start Ollama, click it and wait for the local server to become available.
- If no model is installed, click Download followed by the model name. Zush downloads its suggested model and shows the progress. You can use Cancel to stop the download.
- Wait for the download and connection check to finish. Zush selects the downloaded model and enables Ollama mode when setup succeeds.
- Open AI Rename and test a few representative files before using the setup on a large batch or monitored folder.
The Download button shows the model currently suggested by Zush.
If Ollama is already set up
Section titled “If Ollama is already set up”When Ollama is running and models are installed, Zush shows the Model picker instead of the first-model download. Choose an installed vision-capable model, refresh the list if needed, and use Test to check the connection. Confirm that Ollama is the selected AI mode.
The default Ollama host is http://127.0.0.1:11434. Keep it unless you intentionally configured another server. The Start button is for a local Ollama installation; a custom server address must already be reachable.
Troubleshooting
Section titled “Troubleshooting”| What you see | What to do |
|---|---|
| Download Ollama | Install Ollama from the page opened by the button, then return to Zush. |
| Start Ollama | Click the button, or open the Ollama app yourself, then return to Zush. |
| Download followed by a model name | Ollama is running but has no installed models. Use the button to download the suggested model. |
| Model download fails | Check the connection and free disk space, then try the download again. On Mac, use Try again when shown. |
| An installed model is missing from the picker | Refresh the model list and check the host address. The optional Terminal checks below can also help. |
| Test fails | Make sure Ollama is running and that the host and selected model are correct. |
| Analysis is slow or inaccurate | Test a smaller or different vision model and check its memory requirements. |
Alternative: Terminal setup
Section titled “Alternative: Terminal setup”Use this route if you prefer to manage Ollama yourself or want a model other than the one offered by Zush. The buttons above are enough for the default setup.
-
Install Ollama from the official download page and open it.
-
In Terminal, download a vision-capable model. For example:
Terminal window ollama pull qwen2.5vl:3b -
If the Ollama server is not already running, start it:
Terminal window ollama serve -
Confirm that the model is installed:
Terminal window ollama list -
Return to Zush Settings → AI Modes → Ollama, refresh the model list, select the model, and click Test. Select Ollama mode, then try a small batch.
You do not need to run ollama serve when the Ollama app is already serving the same local endpoint.
Recommended models
Section titled “Recommended models”Start with the model offered by Zush’s Download button. For a different setup, these vision-capable models are alternatives you can install yourself:
| Model | Intended use | Terminal command |
|---|---|---|
qwen2.5vl:3b | Images and screenshots | ollama pull qwen2.5vl:3b |
gemma3:4b | Visual and multilingual content | ollama pull gemma3:4b |
granite3.2-vision:2b | Documents, tables, and charts | ollama pull granite3.2-vision:2b |
You can browse other vision-capable models in the Ollama catalog. Test any model with your own files before automating a large workflow.
Privacy and limits
Section titled “Privacy and limits”With the default local endpoint, supported analysis stays on your computer. Zush does not silently fall back to cloud AI while Ollama mode is selected. Licensing, update, and other operational checks may still use the network.
Local results depend on the model and the evidence Zush can extract. See Audio and video analysis for recordings: local models do not receive raw audio for cloud-style transcription.