🛡️ AgentGate — AI Agent Security Firewall & Approval Hub

Security for the Autonomous Era: AgentGate is a real-time security firewall, command center, and approval hub for autonomous AI agents. Intercept, evaluate, and control high-risk agent actions before they execute in the real world.

---

🌟 Live Demo & Deployment

• 🌐 Production App: https://agentgate-henna.vercel.app
• ⚙️ Vercel Console: https://vercel.com/meia/agentgate

---

📖 Overview

As autonomous AI agents (powered by LangChain, AutoGPT, CrewAI, or custom LLM loops) gain access to real-world APIs, databases, and financial endpoints, the risk of unmonitored actions increases exponentially.

AgentGate acts as an inline security gateway between your AI agents and external tools:

1. Interception: Agents submit proposed actions to AgentGate before executing them.
2. AI Evaluation: Built-in Sentry AI Evaluators stream risk assessments and determine whether an action is safe or high-risk.
3. Human-in-the-Loop: High-risk actions are paused and routed to an interactive Approval Queue for human verification.
4. Audit Trail: Every attempt, risk score, approval, and rejection is recorded in immutable database logs.

---

🏗 System Architecture

flowchart TD
    subgraph Autonomous Environment
        A[Python / LangChain / CrewAI Agent]
    end

    subgraph AgentGate Security Platform
        B[Supabase Edge Function: submit-proposal]
        C{Sentry AI Evaluator}
        D[Approval Queue Dashboard]
        E[(Supabase Database: audit_log)]
        F[Supabase Edge Function: tool-executor]
    end

    subgraph External Systems
        G[REST APIs / Databases / SendGrid / Stripe]
    end

    A -->|1. Submit Proposed Action| B
    B -->|2. Stream Action Context| C
    C -->|Auto-Approved: Low Risk| F
    C -->|High Risk Flagged| D
    B -->|3. Log Decision Request| E
    D -->|4. Human Manager Approves / Rejects| E
    F -->|5. Safe Execution| G

Loading

---

✨ Key Features

• 🛡️ Real-Time Action Interception: Intercept dangerous actions (e.g. mass marketing blasts, budget changes, schema alterations) before execution.
• 🧠 Sentry AI Risk Evaluator: Server-sent events (SSE) stream step-by-step reasoning from evaluator LLMs analyzing action risk.
• 🚦 Human Approval Queue: Clean, intuitive interface for security admins to review risk justifications, inspect parameters, and approve or reject pending requests.
• 📊 Analytics Dashboard: Real-time stats on total active agents, pending proposals, trust scores, and system health metrics.
• 🤖 AI Agent Builder & Wizard: Prompt-based generator that crafts custom agent specifications, system prompts, and security policies.
• 🔌 Tools & Data Sources Registry: Manage REST read/write tools, database connections, and Bright Data zone configurations.
• 📜 Comprehensive Audit Logging: Filterable, immutable ledger of all agent activities and human interventions.
• 🐍 Python SDK & API Integration: Simple REST endpoint integration compatible with any Python LLM framework.

---

🛠 Tech Stack

Frontend

• Framework: React 18 + Vite 7
• Language: TypeScript
• Styling: Tailwind CSS v4 (Vanilla CSS tokens & Glassmorphism design system)
• Routing: React Router v7
• Animations: Framer Motion
• Icons: Lucide React
• Charts: Recharts

Backend & Storage

• Database: Supabase PostgreSQL with Row Level Security (RLS)
• Serverless Functions: Supabase Edge Functions (Deno / TypeScript)
• AI Provider Integrations: AIML API / OpenAI / Anthropic compatible endpoints

---

📁 Repository Structure

AgentGate/
├── index.html                   # HTML Entry point
├── package.json                 # Frontend dependencies & scripts
├── tsconfig.json                # TypeScript configuration
├── vite.config.ts               # Vite build configuration
├── vercel.json                  # Vercel SPA routing rules
├── .vercelignore                # Vercel deployment exclusions
├── demo_agent.py                # Python SDK integration test script
├── requirements.txt             # Python dependencies for demo script
├── public/                      # Static assets & icons
├── src/
│   ├── App.tsx                  # Client router & navigation structure
│   ├── main.tsx                 # React entry point
│   ├── index.css                # Master Tailwind & glassmorphism theme
│   ├── components/              # Core UI components & step wizards
│   ├── context/                 # Application context providers
│   ├── constants/               # Global configuration constants
│   ├── lib/                     # Supabase client & Edge Function helpers
│   └── pages/                   # Application views
│       ├── LandingPage.tsx      # High-impact Hero landing page
│       ├── Dashboard.tsx        # Command Center dashboard
│       ├── ApprovalQueue.tsx    # Human approval management
│       ├── AuditLog.tsx         # Immutable decision logs
│       ├── Agents.tsx           # Active agent list & trust scores
│       ├── BuilderAgents.tsx    # Agent builder hub
│       ├── AgentBuilderWizard.tsx # Multi-step agent setup
│       ├── AgentDetail.tsx      # Single agent breakdown & telemetry
│       ├── BuilderTools.tsx     # Tool integration registry
│       └── BuilderDataSources.tsx # Data source connections
└── supabase/
    ├── migrations/              # Database schema definition files
    └── functions/               # Serverless Edge Functions
        ├── build-agent/         # AI-assisted agent prompt generator
        ├── generate-proposal/   # Mock proposal generator for testing
        ├── resolve-approval/    # Approval status update handler
        ├── run-agent/           # Agent execution loop
        ├── sentry-orchestrator/ # Multi-agent AI risk evaluation stream
        ├── submit-proposal/    # External agent action interception endpoint
        └── tool-executor/      # Controlled REST tool runner

---

🚀 Getting Started

Prerequisites

• Node.js: v18.0.0 or higher
• npm: v9.0.0 or higher
• Python: 3.9+ (optional, for running demo_agent.py)

1. Clone & Install

# Clone the repository
git clone https://github.com/ranazain9/AgentGate.git
cd AgentGate/AgentGate

# Install dependencies
npm install

2. Configure Environment Variables

Create a .env file in the project root based on .env.example:

VITE_SUPABASE_URL=https://<your-supabase-ref>.supabase.co
VITE_SUPABASE_ANON_KEY=<your-supabase-anon-key>

3. Run Development Server

npm run dev

Open http://localhost:5173 in your browser to launch AgentGate.

4. Build for Production

npm run build

The output bundle will be generated inside the dist/ directory.

---

🐍 Python Agent Integration Example

Autonomous agents can request permission from AgentGate via a simple HTTP POST request:

import requests

AGENTGATE_WEBHOOK = "https://<your-supabase-ref>.supabase.co/functions/v1/submit-proposal"
API_KEY = "your-agent-api-key"

response = requests.post(
    AGENTGATE_WEBHOOK,
    headers={
        "Authorization": f"Bearer {API_KEY}",
        "Content-Type": "application/json"
    },
    json={
        "agentName": "marketing-agent",
        "action": "Execute 50% discount blast to 850,000 users via SendGrid API",
        "riskJustification": "Boost Q3 sales despite exceeding daily promo limit of $500"
    }
)

result = response.json()

if result.get("status") == "APPROVED":
    print("✅ Auto-approved by Sentry AI! Executing action...")
elif result.get("status") == "PENDING_APPROVAL":
    print("⏳ High risk detected! Action routed to AgentGate Approval Queue.")

Run the provided demonstration script:

pip install -r requirements.txt
python demo_agent.py

---

🌐 Deploying to Vercel

Option 1: Vercel CLI (Quickest)

npx vercel --prod

Option 2: Vercel Dashboard (GitHub Integration)

1. Import your GitHub repository (ranazain9/AgentGate) in Vercel.
2. Set Root Directory to AgentGate (if deploying from a nested folder).
3. Configure the required Environment Variables:
   • VITE_SUPABASE_URL
   • VITE_SUPABASE_ANON_KEY
4. Click Deploy.

---

📜 License & Acknowledgments

This project was built for the AI Factory Hackathon presented by lablab.ai.

Distributed under the MIT License. See LICENSE for more information.