Feature request: Provide remote server config

> hf agents run pi --help
> Error: 'llama-server' is required. Please install llama.cpp:
> https://github.com/ggml-org/llama.cpp/blob/master/docs/install.md

This project seems to be essentially openclaw, broken down into the individual pi agent, and a way to run it. That much is great, much more individual control.

However, openclaw can run a single agent, so this becomes useful when you DO want to run multiple agents, but isolated.

IF that's the intent, then it seems to make little sense for each one to run its own llama.cpp server instance. That would be super isolated, but super-inefficient: a centralised vllm or sglang instance that can serve many requests concurrently, or even a llama.cpp instance that can at least make use of good centralised hardware and serve concurrent requests reasonably well is a much better fit.