I've started llama.cpp as a web service like explained here: https://github.com/ggerganov/llama.cpp/blob/master/examples/server/README.md, so I can request with a simple curl command:

    --url <http://localhost:8080/completion> \
    --header "Content-Type: application/json" \
    --data '{"prompt": "Building a website can be done in 10 simple steps:","n_predict": 128}'```