Ollama (Self-Hosted)

Ollama runs open models on your own hardware, so review text never leaves your infrastructure.

1. Install Ollama and pull a model

On the server that will run the model:

curl -fsSL https://ollama.com/install.sh | sh
ollama pull llama3.2

Any chat model works. A small model is enough for a 120-word summary.

2. Make it reachable from Magento

The Magento server calls Ollama directly, so the endpoint must be reachable from the Magento server, not from your browser.

# run on the Magento server
curl http://localhost:11434/api/tags

A JSON list of models means it is reachable. If Ollama runs on another host, start it with OLLAMA_HOST=0.0.0.0 and open port 11434 to the Magento server only.

3. Connect it

In Stores → Configuration → Webkul → AI Product Review Summary:

  1. Select Ai model → Ollama.
  2. Ollama Endpoint → http://localhost:11434 (or your server's URL).
  3. Is Authorized → Yes only if a reverse proxy in front of Ollama requires a bearer token; then fill in Enter Api Key.
  4. Validate Key and Load Models, pick a model, Save Config.

Ollama endpoint validated without authorization

How generation calls Ollama

POST {endpoint}/api/generate
{"model": "<AI Model>", "prompt": "<summary prompt>", "stream": false}

The request times out after 60 seconds. On slow hardware with many long reviews, use a smaller model.

Ollama errors

MessageFix
Ollama Endpoint is missing.Fill in Ollama Endpoint and save.
Ollama model is not selected.Validate and choose an AI Model.
Unable to fetch models from Ollama endpoint.Endpoint unreachable from the Magento server.
Unable to connect to Ollama. Please check endpoint and try again.Ollama is down or the URL is wrong.
Unable to fetch summary from Ollama endpoint.Ollama answered but returned no summary. Check its logs.

Next: Switching Providers.