Documentation and configuration files for a fully local setup

Is your feature request related to a problem? Please describe.
While cloud-hosted large models provide performance that cannot me matched locally due to infrastructure cost or the model unavailability, having this kind of external connectivity requirement for robots can be a no-go for some applications. Even though it is possible, it is currently not clear how to make RAI work fully locally, and what are its platform requirements.

Local does not necessarily mean on-board, as the possibilities would be too limiting at the moment.

Describe the solution you'd like
Documentation pages and ready configuration files for a fully local RAI setup, clearly stating requirements for GPU memory and other platform parameters. If this depends on a selection of models, give numbers for a few configurations that makes sense and cover lower, mid, and higher end. Indicate the trade-offs such as lower task performance overall.

Look at packages such as llama_ros, whether they can be directly used with the project.

Describe alternatives you've considered
There is no alternative that resolves the problem.

Additional context
See https://github.com/mgonzs13/llama_ros

Start by surveying the repository's existing documentation and configuration entry points, then evaluate whether llama_ros can support a fully local setup. Document platform and GPU-memory requirements, provide configurations for lower-, mid-, and higher-end models, and describe the performance trade-offs; the work is done when a user can follow the documented setup without cloud connectivity.