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.