abhibro936/openenv-quantumark-finops

☁️ Cloud FinOps & Security Simulator

An interactive OpenEnv simulation for autonomous agents to optimize cloud infrastructure costs and resolve critical security incidents.

  

📋 Overview

This simulator models a realistic cloud environment with EC2 instances, Elastic IPs, IAM users, and security alerts. Autonomous agents must act as Cloud FinOps and Security engineers to:

• —✅ Detect and eliminate cost waste
• —✅ Resolve critical security breaches
• —✅ Maintain production stability

Perfect for: Training and evaluating LLM-based agents on real-world cloud operations scenarios.

🎮 Three Escalating Tasks

🏗️ Architecture

Environment State (Observation)

{
  "ec2_instances": [
    {
      "instance_id": "i-webapp-prod",
      "instance_type": "t3.large",
      "state": "running",
      "cpu_utilization_percent": 65.0,
      "monthly_cost": 60.0
    }
  ],
  "elastic_ips": [
    {
      "allocation_id": "eip-001",
      "public_ip": "203.0.113.1",
      "associated_instance_id": null,
      "monthly_cost": 3.65
    }
  ],
  "iam_users": [
    {
      "user_name": "dev-john",
      "has_mfa_enabled": false,
      "active_access_keys_count": 2
    }
  ],
  "security_alerts": [
    {
      "alert_id": "sec-alert-001",
      "severity": "critical",
      "description": "Compromised credentials suspected for dev-john"
    }
  ],
  "billing_summary": {
    "total_monthly_spend": 4194.60,
    "projected_monthly_spend": 5000.0,
    "budget_limit": 1000.0
  }
}

Available Actions

terminate_instance(instance_id)         # Stop running instances
downsize_instance(instance_id, new_type) # Right-size to cheaper types
release_elastic_ip(allocation_id)       # Remove unattached EIPs
revoke_iam_key(user_name, access_key_id) # Disable compromised keys
no_op()                                 # Valid no-operation action

🚀 Quick Start

Option 1: Try Live (Hugging Face Space)

Visit: [https://huggingface.co/spaces/abhibro936/openenv-quantumark-finops](https://huggingface.co/spaces/abhibro936/openenv-quantumark-finops)

All tasks are instantly accessible via REST API. No setup required.

Option 2: Run Locally

Prerequisites

• —Python 3.10+
• —Docker (optional, for containerized deployment)

Installation

# Clone the repository
git clone https://github.com/ABHISHEK-DBZ/Openev-qua-tumark-finops.git
cd Openev-qua-tumark-finops

# Create virtual environment
python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

Start the Server

# Terminal 1: Start FastAPI server
uvicorn main:app --reload --host 0.0.0.0 --port 7860

# Now accessible at: http://localhost:7860

Run an Agent

# Terminal 2: Execute agent inference
export API_BASE_URL="http://localhost:11434/v1"  # Ollama or OpenAI endpoint
export MODEL_NAME="llama3.2"
export HF_TOKEN="your-token-or-ollama"
export ENV_BASE_URL="http://localhost:7860"

python inference.py

📡 API Reference

Initialize Task

POST /reset?task_id=1

Response: Initial observation JSON for the specified task.

Get Current State

GET /state?task_id=1

Response: Current environment state.

Execute Action

POST /step?task_id=1
Content-Type: application/json

{
  "action_type": "release_elastic_ip",
  "allocation_id": "eip-001"
}

Response:

{
  "observation": {...},
  "reward": 0.33,
  "done": false,
  "info": {}
}

View Dashboard

GET /

Response: HTML dashboard with task stats and live metrics.

🐳 Docker Deployment

Build Image

docker build -t cloud-finops-sim:latest .

Run Container

docker run -it -p 7860:7860 \
  -e API_BASE_URL="http://host.docker.internal:11434/v1" \
  -e MODEL_NAME="llama3.2" \
  -e HF_TOKEN="ollama" \
  cloud-finops-sim:latest

📊 Scoring & Grading

Each task has independent scoring (0.0–1.0 range):

Task 1: EIP Cleanup

• —Release each orphaned EIP: +0.33
• —Release attached EIP (eip-004): -0.30 penalty
• —Max Score: 1.0

Task 2: Database Rightsizing

• —Exact target (t3.medium): 1.0
• —Any downsize attempt: 0.5
• —No action: 0.0

Task 3: Security Incident

• —Each rogue terminated: 0.25
• —Revoke compromised keys: 0.5
• —Terminate production instance: -0.5 penalty
• —Max Score: 1.0

📝 Inference Script Format

Your agent must emit strict stdout format for evaluation:

[START] task=Task Name env=Cloud FinOps & Security Simulator model=llama3.2
[STEP] step=1 action={'action_type': 'no_op'} reward=0.00 done=false error=null
[STEP] step=2 action={'action_type': 'release_elastic_ip', 'allocation_id': 'eip-001'} reward=0.33 done=false error=null
[END] success=true steps=2 score=1.00 rewards=0.00,0.33

Critical Format Rules:

• —[START] must appear once per task
• —[STEP] emitted after each action
• —[END] must close the task
• —Rewards formatted to 2 decimal places
• —No newlines in action dict strings

🏗️ Project Structure

.
├── main.py              # FastAPI server & endpoints
├── inference.py         # LLM agent loop (entrypoint)
├── requirements.txt     # Python dependencies
├── Dockerfile          # Container definition
├── openenv.yaml        # OpenEnv spec (3 tasks)
├── src/
│   ├── env.py         # CloudFinOpsEnv class
│   └── models.py      # Pydantic request/response models
└── tests/
    └── test_env.py    # Unit tests

🔧 Configuration

Set environment variables before running:

📚 OpenEnv Compliance

✅ Full OpenEnv 0.1 Specification Compliance:

• —3 independent tasks with graders
• —Deterministic step/reset/state interface
• —JSON observation & action schemas
• —Reward in [0.0, 1.0] range
• —done flag for episode termination

🤝 Contributing

1. 1.Fork the repo
2. 2.Create a feature branch (git checkout -b feature/improvement)
3. 3.Commit changes (git commit -m "Add improvement")
4. 4.Push to GitHub (git push origin feature/improvement)
5. 5.Open a Pull Request

📄 License

MIT License — See LICENSE file for details.

🎯 Use Cases

• —LLM Fine-tuning: Train agents on realistic cloud operations
• —Agentic Benchmarking: Evaluate autonomy & decision-making
• —FinOps Training: Educational sandbox for cost optimization
• —Security Response Drills: Practice incident response at scale

🔗 Resources

• —OpenEnv GitHub
• —CloudFinOps Best Practices
• —AWS Cost Optimization

Made with ❤️ by [@ABHISHEK-DBZ](https://github.com/ABHISHEK-DBZ) [@Paritosh2681](https://github.com/Paritosh2681)