NonStop Operations Assistant

Reference implementation for the CAIL challenge. Translates plain-English requests into HPE NonStop commands, and refuses to guess when it doesn't have enough to go on.

Internal prototype for review by Speer and CAIL. Not the student-facing version.

Internal draft

What this demonstrates. The hard part isn't generating a command. It's knowing when the request is underspecified or carries a risk, and saying so before the operator hits enter. Every response here puts reasoning and warnings above the command, never below it and never suppressed.

Operator request

Try these

Show naive output too — what the current prototype returns when you ask it for the command only, with no explanations and no warnings.

How it works. Each request is matched against a small knowledge base of NonStop commands. Every entry declares its required parameters, why each one matters, and any caveats that apply. Missing a required parameter blocks the command and produces a question instead. A caveat produces a warning that sits above the command.

Where the LLM goes. This prototype uses deterministic matching so it demos reliably offline. In a real build, a model does intent classification and parameter extraction, with retrieval over the manuals supplying the parameter requirements and caveats. The structure of the response stays exactly as it is here: the model proposes, the knowledge base constrains, the operator approves.

Honesty about syntax. Some entries in this knowledge base are marked low-confidence, and the assistant says so rather than presenting a guess as fact. That's the behaviour being demonstrated, not a limitation to work around.