Show HN: An AI concierge that never auto-executes actions (action-confirm only)

I built an AI-first system that separates “conversation” from “decision”.

Instead of letting AI directly execute actions, the system only suggests actions. Every operation requires explicit action confirmation.

Key design rules: - No “main language” fallback (13 languages treated equally) - AI never auto-executes actions - A dedicated Trust Decider layer to keep behavior deterministic under ambiguity - Designed and built solo, running on a shared host

This repository is the clean core extracted from a real production system.

I’m sharing this to get feedback from people who’ve shipped AI systems in production. I’m also open to short-term consulting or contract work if this resonates.


Comments URL: https://news.ycombinator.com/item?id=46345070

Points: 1

# Comments: 0