Vienna OS

The execution kernel for AI agents.

Agents propose. Vienna OS decides. Every action warranted. Every execution verified.

Website · Console · Docs · Try Live

---

The Problem

AI agents are taking real-world actions. A smarter model is not the answer to:

• Network timeouts that duplicate a $50K transaction
• Stale context from inference latency causing wrong decisions
• Prompt injections that bypass "please be careful" guardrails
• Zero audit trail for regulators asking "why did the AI do that?"

These are infrastructure problems, not intelligence problems.

The Solution

Vienna OS is the deterministic runtime boundary between agent reasoning and real-world side effects. Like an OS kernel validates every syscall before allowing state modification, Vienna OS validates every agent intent before allowing execution.

Agent → Intent → Policy Eval → Risk Tier → Warrant → Execute → Verify → Audit

Prompts are not permissions. Agent output is untrusted input. Every action is a proposal — a claim the agent wants to act, not an order that it will.

Quick Start

Node.js

npm install vienna-os

import { ViennaClient } from 'vienna-os';

const vienna = new ViennaClient({
  baseUrl: 'https://console.regulator.ai',
  agentId: 'your-agent-id',
});

const result = await vienna.submitIntent({
  action: 'deploy',
  payload: { service: 'api-gateway', version: 'v2.4.1' },
});

// result.pipeline: "executed" | "pending_approval" | "denied" | "simulated"
// result.warrant: { id, signature, expires_at } — cryptographic execution authority

Python

pip install vienna-os

from vienna_os import ViennaClient, Intent

vienna = ViennaClient(
    base_url="https://console.regulator.ai",
    agent_id="your-agent-id",
)

result = vienna.submit_intent(Intent(
    action="deploy",
    payload={"service": "api-gateway", "version": "v2.4.1"},
))

REST API

curl -X POST https://console.regulator.ai/api/v1/agent/intent \
  -H "Content-Type: application/json" \
  -d '{
    "agent_id": "your-agent-id",
    "action": "deploy",
    "payload": {"service": "api-gateway"}
  }'

Architecture

Vienna OS implements the Decision Intelligence Runtime pattern:

Pillar	What it does
Policy as a Claim	Agent output is untrusted input. Every action is a proposal, not an order.
Contracts as Code	Deterministic policy evaluation — not another LLM asking "is this dangerous?"
Cryptographic Warrants	SHA-256 signed, time-limited, scope-constrained execution authority.
JIT Verification	Context drift detection before execution. Stale decisions are aborted.
Idempotency	Unique execution keys prevent duplicate transactions on retry.
Decision Flow Tracing	One correlation ID reconstructs the entire decision chain.

Risk Tiers

Tier	Approval	Examples
T0	Auto-approve	Read-only queries, health checks, status
T1	Single operator	Config changes, service restarts, data writes
T2	Multi-party	Deployments, payments, data deletion
T3	Justification + rollback	Production DB migrations, financial transactions

Console

Live at console.regulator.ai:

• Now — Live system posture dashboard
• Intent — Submit actions through the pipeline
• Approvals — Approve/deny pending proposals
• Fleet — Agent registry with trust scores
• Policies — Policy-as-code rule management
• History — Immutable audit trail
• Compliance — One-click governance reports

Framework Agnostic

Works with any system that makes HTTP requests:

• OpenClaw
• LangChain / LangGraph
• CrewAI
• AutoGen
• Custom agents

5 lines to integrate. One API.

Pricing

Plan	Price	Agents
Community	Free	5
Team	$49/agent/mo	25
Business	$99/agent/mo	100
Enterprise	Custom	Unlimited

License

Business Source License 1.1 — free for evaluation and non-production use. Converts to Apache 2.0 on 2030-01-01.

Built By

ai.ventures — Cornell Law × systems engineering. Patent pending: USPTO #64/018,152.

---

Prompts are not permissions.