Show HN: Constitutional AI Agent OS (governance enforced at kernel level) (github.com)

A new open-source “Constitutional AI” stack — steward-protocol — was released as a VibeOS cartridge pack that enforces cryptographic governance at the kernel level for autonomous agents. Branded as “Artificial Governed Intelligence (A.G.I.),” the package implements a Steward Protocol that gives each agent a verifiable ECDSA identity, signs every action, and records an append‑only, cryptographically-signed event ledger (embedded SQLite: data/vibe_ledger.db). Agent City is the reference deployment: seven cartridges (HERALD, CIVIC, FORUM, SCIENCE, ARCHIVIST, ARTISAN, ENVOY) that provide governance, auditing, NAT language interface, and media ops; system state and authoritative snapshots are exposed via vibe_snapshot.json and kernel commands. Authors claim production readiness, self-healing, and easy installation into VibeOS (kernel >=2.0), plus test tooling (pytest). For the AI/ML community this is significant because it moves identity, accountability and governance from application-layer policy into the OS/kernel layer for agent ecosystems — enabling cryptographic provenance, audit trails, on‑chain‑style immutability, and enforceable proposal/voting workflows for autonomous workflows. Key technical implications: non-repudiable signatures for every action (keys stored locally, auto-generated), immutable append-only event logs for reproducibility, and a protocol-first governance model that can reduce “black box” trust claims. Practical considerations remain: host key protection, single-node SQLite scalability, threat model for kernel compromise, and how policy rules map to model behavior. The repo provides architecture docs, a constitution, and example integration points for developers to experiment with governed agent deployments.