Skip to main content
🔒 Transparent AI Governance Real-time trust metrics from the Vorion platform

System Architecture

How Vorion's trust engine, governance framework, and edge dashboard work together to provide transparent, real-time AI agent oversight.

Platform Packages

A

ATSF Core

Agent Trust & Safety Framework — the runtime engine for trust scoring and governance.

  • Trust Engine — score computation + tier promotion
  • Governance Engine — rule evaluation + decisions
  • Sandbox Training — 21-challenge boot camp
  • Fastify API — REST endpoints
P

Platform Core

Enterprise platform services — authentication, registry, and cryptographic proof.

  • Auth — MFA (TOTP) + PIV/CAC smart card
  • Agent Registry — tenant + API key management
  • Proof Chain — hash-chained signed records
  • HSM — PKCS#11 hardware security module
C

Contracts

Shared type definitions, database schemas, and canonical enums across all packages.

  • Drizzle ORM schemas (PostgreSQL)
  • TrustTier, ControlAction enums
  • Intent, Decision, Agent types
  • BASIS + CAR specification types

Edge Dashboard Layer

The public dashboard runs on Cloudflare's edge network. A scheduled Worker syncs data from PostgreSQL every 5 minutes into KV, which Astro pages read at request time. This architecture keeps database credentials off the edge and serves cached data in <50ms globally.

🗄️

Neon PostgreSQL

Source of truth — agents, intents, policies, trust scores

⚡

Scheduled Worker

5-min cron — syncs stats, health checks, npm, CI

📦

Cloudflare KV

Edge cache — 300s TTL, 7 data keys

🌐

Astro SSR

Pages + API routes — reads from KV at the edge

8-Tier Trust Model

Every agent starts at T0 (Sandbox) and progresses through tiers as trust signals accumulate. Scores range from 0 to 1000. Higher tiers unlock greater autonomy.

0 200 350 500 650 800 876 951 1000
T0 Sandbox

0 – 199

T1 Observed

200 – 349

T2 Provisional

350 – 499

T3 Monitored

500 – 649

T4 Standard

650 – 799

T5 Trusted

800 – 875

T6 Certified

876 – 950

T7 Autonomous

951 – 1000

Data Flow Pipelines

Trust Scoring Pipeline

1 Agent registers + enters Sandbox (T0)
2 Boot Camp — 21 adversarial challenges
3 Challenge results → Trust Signals
4 Trust Engine computes score → Tier promotion
5 Graduation → T0 → T1 transition

Governance Pipeline

1 Agent submits action request (Intent)
2 Governance Engine evaluates against rules
3 Decision: approve / escalate / block
4 Proof Bridge creates signed record
5 Proof Chain — hash-linked, verifiable audit trail

Technology Stack

Runtime

Node.js + TypeScript

Database

PostgreSQL (Neon) + Drizzle ORM

API

Fastify + REST

Edge

Cloudflare Workers + KV + Astro

Monorepo

Turborepo + pnpm workspaces

Auth

TOTP MFA + PIV/CAC + HSM

Testing

Vitest (398+ tests)

CI/CD

GitHub Actions + Wrangler