This write-up documents my settings for FireFox "AI. The advantage of the native solution over, e.g., Page Assist is that any text that can be manually selected can be fed to a local LLM, even if the text can't be printed (e.g., articles in SEP).
For integration of a local LLM into the FireFox sidebar a web UI is loaded into a frame.
Features thus depend on the UI elements. While llama-server provides a suitable element, vLLM can only be used when served through, e.g., Open Web-UI.
When llama-server is used behind llama-swap the upstream path must be used.

