Install Ollama
Ollama runs an open-weights model on your own hardware. Reviews never leave your infrastructure and there is no per-token bill. In exchange you provide the machine and accept that a local model is usually slower than a hosted API.
1. Install the server
On the machine that will host it:
curl -fsSL https://ollama.com/install.sh | sh
2. Pull a model
ollama pull gpt-oss
ollama list
Any chat-capable model works — gpt-oss, gemma3, granite4, and so on. Translation quality varies a lot between small models, so test on real reviews before committing. See Choosing a Model.
3. Sizing the machine
| Model size | Practical minimum |
|---|---|
| 3–4B | 8 GB RAM, CPU workable |
| 7–12B | 16 GB RAM, GPU strongly preferred |
| 30B+ | GPU with 24 GB+ VRAM |
CPU-only inference works but is slow enough that you will need a much higher request timeout — see Timeouts & Batch Size.
4. Make it reachable
If Magento and Ollama share a machine, the default http://localhost:11434 is enough. Otherwise bind it to an interface Magento can reach:
OLLAMA_HOST=0.0.0.0:11434 ollama serve
Do not expose Ollama to the internet
Plain Ollama has no authentication. Keep it on a private network, or behind a reverse proxy that adds a bearer token — then set Is Authorized to Yes in the admin.
Next: Connect Ollama to Magento.