aluminatiai 0.2.0

GPU energy monitoring agent — per-job cost attribution for AI teams

AluminatAI - GPU Energy Intelligence Platform

Know exactly what your GPUs cost. Every watt, every dollar, every job.

AluminatAI is an open-source GPU energy monitoring platform that gives AI teams real-time visibility into power consumption, energy costs, and utilization across their GPU fleet. A lightweight Python agent runs on your GPU machines and streams metrics to a cloud dashboard where you can track spending, compare jobs, and optimize workloads.

Live: https://www.aluminatiai.com/

---

How It Works

┌──────────────────┐     HTTPS/JSON      ┌──────────────────────────┐
│   GPU Machine     │ ─────────────────► │   AluminatAI Platform     │
│                    │   every 60s        │                            │
│  ┌──────────────┐ │                     │  ┌──────────┐             │
│  │  Python Agent │ │                     │  │ Next.js  │  Vercel     │
│  │  (pynvml)    │ │                     │  │ API      │             │
│  └──────────────┘ │                     │  └────┬─────┘             │
│                    │                     │       │                    │
│  NVIDIA A100/H100  │                     │  ┌────▼─────┐             │
│  RTX 3090/4090     │                     │  │ Supabase │  PostgreSQL │
│  Any NVIDIA GPU    │                     │  │ Database │  + RLS      │
└──────────────────┘                     │  └────┬─────┘             │
                                          │       │                    │
                                          │  ┌────▼─────┐             │
                                          │  │Dashboard │  React      │
                                          │  │ UI       │  + Recharts │
                                          │  └──────────┘             │
                                          └──────────────────────────┘

Features

• Real-Time GPU Monitoring - Power draw, utilization, temperature, memory, and clock speeds sampled every 5 seconds
• Energy Cost Tracking - Calculates energy consumption in kWh and converts to dollar costs at your electricity rate
• Job Attribution - Track which training jobs consumed how much energy and what they cost
• Dashboard - Three views: Today's Cost, Jobs Table, and Utilization vs Power chart
• Free Trial - 30-day free trial with auto-generated API keys on signup
• Lightweight Agent - <1% CPU, ~50MB RAM overhead on GPU machines
• Secure - Row-Level Security, API key auth with pgcrypto, rate limiting, server-side validation
• Minimax Scheduler - Bonus hackathon project: AI-powered job scheduling that balances speed vs. energy cost

---

Project Structure

AluminatAI/
├── aluminatai-landing/          # Next.js web platform (deployed to Vercel)
│   ├── app/
│   │   ├── api/
│   │   │   ├── metrics/ingest/  # GPU metrics ingestion endpoint
│   │   │   ├── dashboard/       # today-cost, jobs, utilization-chart
│   │   │   ├── user/profile/    # User profile + API key rotation
│   │   │   └── cron/            # Materialized view refresh
│   │   ├── dashboard/           # Protected dashboard UI
│   │   ├── login/               # Auth pages
│   │   └── page.tsx             # Landing page
│   ├── components/              # React components
│   ├── lib/                     # Auth, rate limiting, Supabase clients
│   └── database/migrations/     # SQL migrations (001-005)
│
├── agent/                       # Python GPU monitoring agent
│   ├── main.py                  # Agent entry point
│   ├── collector.py             # NVML-based GPU metrics collector
│   ├── uploader.py              # API upload with retry + local backup
│   ├── config.py                # Environment-based configuration
│   ├── install.sh               # One-line install script
│   └── tests/                   # Test suite + Colab notebook
│
├── minimax-scheduler/           # Hackathon: Minimax GPU job scheduler
│   └── backend/                 # FastAPI + minimax algorithm
│
├── backend/                     # Legacy FastAPI backend (reference)
├── frontend/                    # Legacy React frontend (reference)
├── docker/                      # Docker configs for agent + backend
├── docs/                        # Architecture docs, metrics schema
└── assets/                      # Logo and diagrams

---

Quick Start

Prerequisites

• Node.js 18+ and npm
• Python 3.8+
• A Supabase account (supabase.com)
• An NVIDIA GPU (for the agent) or Google Colab with GPU runtime

1. Clone the Repository

git clone https://github.com/AgentMulder404/aluminatai-landing.git
cd aluminatai-landing

2. Set Up the Database (Supabase)

1. Create a new project at supabase.com
2. Go to SQL Editor and run the migrations in order:

# Run these SQL files in the Supabase SQL Editor:
database/migrations/002_gpu_monitoring_schema_postgres.sql
database/migrations/003_fix_materialized_view.sql
database/migrations/004_fix_trigger_permissions.sql
database/migrations/005_secure_api_keys_and_constraints.sql

This creates:

• users table with auto-generated API keys (using pgcrypto)
• gpu_metrics time-series table with CHECK constraints
• gpu_jobs table for job tracking
• gpu_metrics_hourly materialized view for fast dashboard queries
• Row-Level Security policies on all tables
• Triggers for user profile auto-creation on signup

3. Set Up the Web Platform

cd aluminatai-landing
npm install

Create a .env.local file:

# Supabase (from your project settings > API)
NEXT_PUBLIC_SUPABASE_URL=https://your-project.supabase.co
NEXT_PUBLIC_SUPABASE_ANON_KEY=your-anon-key
SUPABASE_SERVICE_ROLE_KEY=your-service-role-key

# Cron secret (generate with: openssl rand -base64 32)
CRON_SECRET=your-cron-secret

Run the development server:

npm run dev

Visit http://localhost:3000 - you should see the landing page.

4. Create an Account

1. Click "Start Free Trial" on the landing page
2. Enter your name, email, and password
3. You'll be redirected to the dashboard setup page
4. Copy your API key (starts with alum_)

5. Install the GPU Agent

On your GPU machine (or Google Colab):

# Install dependencies
pip install pynvml requests python-dotenv rich

# Set environment variables
export ALUMINATAI_API_KEY="alum_your_key_here"
export ALUMINATAI_API_ENDPOINT="http://localhost:3000/api/metrics/ingest"

# Run the agent
python agent/main.py

Options:

# Custom sampling interval (1 second)
python agent/main.py --interval 1

# Save to CSV + upload
python agent/main.py --output data/metrics.csv

# Run for 5 minutes
python agent/main.py --duration 300

# Quiet mode (no console output)
python agent/main.py --quiet --output data/metrics.csv

For production, use the systemd service:

cd agent
chmod +x install.sh
sudo ./install.sh

6. Test on Google Colab (A100)

Upload agent/tests/AluminatAI_A100_Test.ipynb to Google Colab:

1. Go to colab.research.google.com
2. File > Upload notebook and select the .ipynb file
3. Runtime > Change runtime type > select A100 GPU
4. Paste your API key in Cell 2
5. Runtime > Run all

The notebook runs 7 test suites:

• NVML hardware access
• Collector class + energy calculation
• API authentication validation
• End-to-end collect + upload
• Stress test under GPU load (8192x8192 matmul)
• API key security audit
• 60-second continuous monitoring demo

---

API Reference

Metrics Ingestion

POST /api/metrics/ingest
Header: X-API-Key: alum_your_key_here

Request body (single metric or array):

[
  {
    "timestamp": "2026-02-06T12:00:00Z",
    "gpu_index": 0,
    "gpu_uuid": "GPU-abc123",
    "gpu_name": "NVIDIA A100-SXM4-40GB",
    "power_draw_w": 285.5,
    "power_limit_w": 400.0,
    "energy_delta_j": 571.0,
    "utilization_gpu_pct": 95,
    "utilization_memory_pct": 60,
    "temperature_c": 72,
    "memory_used_mb": 32000,
    "memory_total_mb": 40960
  }
]

Validation rules:

• power_draw_w: 0-1500W
• temperature_c: 0-120C
• utilization_*_pct: 0-100
• timestamp: valid ISO 8601, not more than 5 minutes in the future
• Max 1000 metrics per request

Rate limit: 100 requests/minute per user

Dashboard APIs

Endpoint	Method	Auth	Rate Limit	Description
/api/dashboard/today-cost	GET	Session	60/min	Today's energy cost
/api/dashboard/jobs	GET	Session	60/min	Job history with pagination
/api/dashboard/utilization-chart	GET	Session	60/min	Time-series chart data
/api/user/profile	GET	Session	-	User profile + API key
/api/user/profile	PATCH	Session	-	Update profile settings
/api/user/profile	POST	Session	5/hr	Rotate API key

API Key Rotation

curl -X POST https://aluminatiai-landing.vercel.app/api/user/profile \
  -H "Content-Type: application/json" \
  -H "Cookie: your-session-cookie" \
  -d '{"action": "rotate_api_key"}'

---

Security

• API Keys: Generated with pgcrypto gen_random_bytes() - 340 bits of entropy
• Row-Level Security: Users can only access their own data
• Rate Limiting: Per-user limits on all endpoints
• Input Validation: Server-side + database CHECK constraints
• HTTPS: Enforced by Vercel
• No ambiguous characters: API keys exclude 0, O, I, l, 1 to prevent copy errors

---

Deployment

Vercel (Web Platform)

# Install Vercel CLI
npm i -g vercel

# Deploy
cd aluminatai-landing
vercel

# Set environment variables in Vercel dashboard

Cron Job (Materialized View Refresh)

Set up a cron job to refresh the hourly metrics view:

• URL: https://your-app.vercel.app/api/cron/refresh-metrics
• Method: POST
• Header: Authorization: Bearer your-cron-secret
• Schedule: Every hour (0 * * * *)

You can use cron-job.org (free) or Vercel Cron.

---

Tech Stack

Component	Technology
Web Framework	Next.js 16
UI	React 19 + Tailwind CSS 4
Charts	Recharts
Database	Supabase PostgreSQL
Auth	Supabase Auth
GPU Agent	Python + pynvml (NVML)
Deployment	Vercel
Scheduler	Minimax with alpha-beta pruning

---

Minimax GPU Scheduler

A bonus hackathon project in minimax-scheduler/ that uses game theory to optimize GPU job scheduling:

• Speed Player (Maximizer): Wants to complete jobs ASAP
• Cost Player (Minimizer): Wants to minimize energy costs
• Alpha-Beta Pruning: Efficiently explores the decision tree
• Result: 15-30% cost savings vs. naive FIFO scheduling

cd minimax-scheduler/backend
pip install -r requirements.txt
python demo.py

---

Contributing

1. Fork the repository
2. Create a feature branch (git checkout -b feature/my-feature)
3. Commit your changes
4. Push to the branch (git push origin feature/my-feature)
5. Open a Pull Request

---

License

This project is open source. See LICENSE for details.

---

Built by @AgentMulder404