Skip to content

Smart RAG: answers from your man pages ​

A model knows commands in general. Your system has the exact documentation for the versions you have installed: its man pages. With Smart RAG, TuxAide answers questions about a specific command from those man pages, and cites them:

text
$ what does the -z flag do in grep
  …
  This command will search for "pattern" in file.txt, treating null
  characters as line separators.

  Source: man grep(1)

Fast when simple, precise when needed ​

TuxAide decides per question. “how do I create a folder” goes straight to the model; “rsync options to exclude files” mentions a command, so the matching man page is looked up first. If the lookup takes too long (rag_timeout, 8 s), the model answers alone.

ModeWhat it does
llmModel only (the default)
smartMan pages when the question needs them (recommended)
deepMan pages for every question, longer answers
bash
tuxaide mode smart

Turning it on ​

  • Installer: answer Y to Smart RAG (needs 5 GB of RAM or more).
  • Any install: tuxaide setup --rag.

TuxAide then installs ChromaDB in its own virtualenv (~/.local/share/tuxaide/venv — never your system Python, no pip needed), downloads the embedding model nomic-embed-text (274 MB), switches to smart, and indexes about 150 common man pages in the background — a few minutes. You can use TuxAide meanwhile; until indexing finishes, answers come from the model alone. tuxaide doctor shows the progress; the log is in ~/.config/tuxaide/logs/index.log.

With pipx, pipx install 'tuxaide[rag]' puts ChromaDB in TuxAide's own pipx environment instead.

Your own commands ​

bash
tuxaide index nginx    # add one man page
tuxaide reindex        # rebuild the whole index (after big system updates)

MIT License