◈NrtVix AIGUARDIAN / STUDIO PHASE 4.3 · GUARDIAN KEY (LOCAL)
LOCAL DEVELOPMENT LAB — This is not a hosted service, real payment tool, or production security certification.
ACCESS YOUR WORKSPACE

Create a workspace

Use a demo email and a unique password you don't use anywhere else.

Passwords are hashed locally. Guardian Key recovery works only after secure device enrollment. Not for public hosting.
WORKSPACE / OVERVIEW

Your workspace

Your agents. Your decisions. Your control.

ISOLATEDDevelopment tenant
Registered agents0Scoped to this workspace
Policy decisions0Within this tenant
Awaiting review0Human approval required
Sandbox executions0No real money or messages
01 / IDENTITY

Connected agents

Add, rotate, disable

No agents registered yet.

02 / EXECUTION

Guarded action lab

Simulation only
Create an agent and submit a demo request. Refund: $15 ALLOW · $75 REVIEW · $500 BLOCK. Database delete: REVIEWABLE only for one synthetic record; never a real deletion.
03 / WORKSPACE GOVERNANCE

Security policy studio

Policy v1

Shape your workspace's security posture. Settings can make checks stricter, never bypass the built-in safety ceiling. Every edit is logged, versioned, and expires open unexecuted approvals.

Financial action control

Daily total counts authorized synthetic refund amounts (UTC). Nothing transfers real funds.

Data & communication

Campaign & destructive actions

Tool estimate budget is separate from refund amount; this lab enforces a minimum $0.03 estimate per action.

Fail-safe change: saving immediately expires pending, approved and reviewable requests that have not run. Existing completed audit records remain.

■
Emergency stop

All agents in this workspace can run within policy.

04 / PRIVATE CONNECTOR CREDENTIALS

Local mock credential vault

Encrypted at rest · demo keys only

Workspace owners can create, rotate and revoke server-generated fake credentials. The secret is encrypted in a private SQLite file and is never shown to agents, the browser or the mock worker. No Stripe, Shopify, email, or real provider API keys are accepted. This is not a cloud KMS.

No demo credentials created in this workspace yet.

Rotation and revocation cancel already-queued mock connector jobs for that action. Historical v1.7 mock dispatch is retained for offline regression compatibility; the v1.8 secure demo requires this vault. The new vault does not enable live provider access.

◇
Connector Security Boundary v1.7 · Authenticated offline mock worker

Every "Run sandbox" button uses the new guarded gateway: agent identity, action, tool, target, amount, units and approval are checked before ONE simulated receipt is recorded. Authorized sandbox receipts are queued for a separate signed mock-connector boundary. A private worker validates short-lived HMAC jobs and replay protection. No external provider is contacted; worker processing is a separate offline demo. Agents with direct credentials outside Guardian remain uncontrolled.

◇
Connector dispatch status · Offline only

Each authorized sandbox receipt creates a server-owned connector outbox item. The local demo dispatcher verifies current agent permissions and policy version before signing a single short-lived mock request. Pending items are NOT live external actions. Queued mock jobs: 0

◇
Smart Exception Review · Restricted sandbox

A valid single-record synthetic database deletion request starts denied. The agent may request an exception; the workspace owner must explicitly review scope and confirm a mock backup. Even after approval, the gateway records only a preview. No SQL DELETE is ever executed. Hard blocks cannot be overridden.

V2.0 / SAFE AGENT ONBOARDING

Agent & connector readiness

Connect a sandbox agent first, set a workspace policy, then use the trusted gateway. Never give Stripe or email provider keys to an AI agent. This panel does not collect, store or verify real credentials.

01 · Register sandbox agent
Scoped key, shown only once
02 · Guardian policy
ALLOW / APPROVAL / BLOCK
03 · Trusted gateway
Single-use mock execution
04 · Provider check
Optional, read-only TEST-mode CLI

Workspace readiness: not checked yet.

Stripe test account connectivity is NOT verified here. The separate optional terminal checker performs only one read-only API request, with an opt-in key entered privately. Provider refunds, messages and live credentials remain disabled in this dashboard.

PHASE 2 / LIVE PLANNER → WORKSPACE

Real AI + Guardian activity

Local pilot · not cloud SaaS

This dashboard already contains the authenticated, workspace-scoped Guardian API. Phase 2 adds a separate optional terminal bridge so a real OpenAI planner's strictly validated DEMO-only request can produce an event in this same workspace. No OpenAI key is entered into the dashboard and no real service tool is callable.

01 / Make an agent
Choose lookup (view_customer_record) or export (export_customer_data). Save its one-time sandbox key locally.
02 / Run terminal bridge
05_PHASE2_CONNECT_AI_TO_DASHBOARD.bat. Choose offline (free) or OpenAI (paid opt-in); key input is hidden.
03 / Review Guardian activity
Click Refresh below. Export requires the signed-in owner to approve; the gateway remains synthetic-only.

Only synthetic events are stored. The dashboard does not independently attest whether an event came from a real model, a mock planner, or an agent client. The bridge's paid live test is separate from the previous disposable v2.2 live-test script.

PHASE 3 / PRIVATE BETA PREPARATION

Local pilot readiness & audit

Not approved for public hosting

This panel checks only your local synthetic workflow. It cannot certify a real cloud product. No OpenAI key, customer data or provider credential is required.

Local readiness not checked. Cloud private beta: NOT READY.

Before inviting real users: hosting isolation, HTTPS, secret management, MFA, verified email, backups, incident response, monitoring, external security testing, privacy/legal review and a production database are still missing or unverified. Never expose the local development server to the Internet.

05 / POLICY DECISIONS

Approval & audit activity

Action / targetAmount / unitsDecisionWhenReview / run
No decisions yet.
◈ NrtVix AI / v2.0 Agent Integration LabSandbox-only receipts · 0 real rows / messages / payments · no external tools