API keys · Local AI
Run AI on your own computer.
The Local engine uses Ollama, a free app that runs AI models on your machine. Nothing is sent to a cloud service, there’s no key and no usage limit. The trade-off is a large download and a computer with enough memory.
Before you start
- Memory — Soritch’s default model needs about 8 GB of free RAM; 16 GB or more total is comfortable. Smaller models run on less.
- Disk space — about 5 GB for the default model.
- Speed — Apple Silicon Macs and PCs with a recent graphics card are fastest. Older computers work but write-ups take longer.
1. Install Ollama
- Download Ollama for your system from ollama.com/download.
- Install and open it. On a Mac it lives in the menu bar; on Windows, in the system tray. It needs to be running whenever you use the Local engine.
2. Download a model
Open Terminal (Mac) or PowerShell (Windows) and run this to download Soritch’s default model:
ollama pull deepseek-r1:8b-0528-qwen3-q4_K_MOn a computer with less memory, a smaller model works too, for example:
ollama pull llama3.2:3bBrowse other models in the Ollama library. Check a model works with:
ollama list3. Select it in Soritch
- Open Settings and scroll to AI engine.
- Set Engine to Local.
- Choose your model in Ollama model. The list shows every model you’ve downloaded.
4. Check it works
Open Analytics. The status near the top should show a green dot with the model name. Run an analysis; the first one can take a little longer while the model loads.
Troubleshooting
- “Ollama is not reachable at localhost:11434” — Ollama isn’t running. Open the Ollama app and try again.
- “No models found” — Ollama is running but has no models. Run the
ollama pullcommand above. - Very slow or out-of-memory errors — switch to a smaller model, or use Groq instead.