AgentEyes Presentation

Click anywhere to start

00:00 / 01:00
AgentEyes
Your MX Creative Console as
AI Agent Control Center
Logitech DevStudio 2026
The Problem
AI Agents Act.
Without Oversight.
No physical controls. No risk visibility. No audit trail.
🚫

No Kill Switch

Can't physically stop an agent mid-action

👁

Blind Trust

Actions happen without risk assessment

⚠️

No Gradients

Binary on/off — no autonomy spectrum

The Solution
Physical Control for
Digital Agents
Dial sets autonomy. Buttons decide. LEDs show risk.
LOGITECH
1
Approve
Reject
Pause
Escalate
How It Works
5 Autonomy Levels
0

Manual

Human approves all

1

Supervised

Low-risk auto-approves

2

Guided

Medium auto-approves

3

Autonomous

Only critical needs human

4

Full Auto

No human gates

Real-Time Risk Scoring
Every Proposal Scored
Low Risk

Check Calendar

Read-only calendar access for today's schedule

2
✓ Auto-Approved
Medium Risk

Send Email

Status update to 23 team members — cannot unsend

5
👤 Human Review
Critical Risk

Wire $50,000

Irreversible financial transfer — dual authorization required

9
🛑 Blocked — Human Required
Architecture
Plugin + Bridge + Overlay
MX Creative Console ─── Logi Actions SDK Plugin
    │
    │ WebSocket (:3002)
    ▼
Bridge Server ───── State Manager + Risk Engine
    │               ├─ Agent Simulator
    │               └─ Scenario Engine
    │
    ▼
Web Overlay ────── Next.js + React + TailwindCSS
                     ├─ Console Simulator (interactive replica)
                     ├─ Proposal Cards + Risk Meter
                     └─ Audit Timeline
📜

Dual-Readable Interface

USPTO App #63/981,053
Dual-Readable Interface System and Method for
Human-Agent Co-Execution in Digital Systems
Filed February 12, 2026
Annotator Runtime Gateway Inspector Compliance
AgentEyes
Physical AI agent oversight.
Built for the MX Creative Console.
Agents read data. Humans read intent.
Invented by Ken Liao · FoodyePay Technology, Inc.