⚡ OpenEnv Compliant · Meta × PyTorch × HuggingFace

ARIA

Adaptive Reward & Incident Architecture

The first OpenEnv RL environment for production incident response.
7 tasks · 14 actions · Curriculum · Dual-agent · Trained Llama-3.1-8B

Try Live API → View GitHub →

Tasks

Actions

∞

Scenarios

0.99

Max Score

Environment Tasks

Eight scenarios of escalating operational complexity

Loading tasks...

ARIA Features

What makes this environment unique

🎓

Curriculum Engine

Tracks agent performance per task with rolling averages. Promotes when mastered (avg > 0.75). Scaffolds with hints when struggling (avg < 0.30). Agents always train at the edge of their capability.

View Status →

⚡

Incident Generator

Procedural incidents from seeds 0–99,999. Six failure modes × eight services × variable noise = infinite unique training scenarios. Same seed always produces the same incident.

Try Generator →

🤝

Dual-Agent Mode

Split observability between two agents. Observer sees logs and alerts. Responder sees metrics and dependencies. Neither can solve the incident alone — they must coordinate via share_finding.

AGENT A: Observer

• alerts, logs

share_finding

AGENT B: Responder

• metrics, deps

View Sessions →

🏆 Leaderboard

Rank	Task	Score	Steps	Status
Loading leaderboard...

Quick Start

# 1. Start an incident curl -X POST https://huggingface.co/space_proxy/arijit-07-devops-incident-response/reset -H "Content-Type: application/json" -d '{"task_id": "easy", "seed": 42}' # 2. Read logs (reward: +0.15) curl -X POST https://huggingface.co/space_proxy/arijit-07-devops-incident-response/step -H "Content-Type: application/json" -d '{"action_type": "read_logs", "service": "payment-service"}' # 3. Diagnose (reward: +0.30) curl -X POST https://huggingface.co/space_proxy/arijit-07-devops-incident-response/step -H "Content-Type: application/json" -d '{"action_type": "diagnose", "root_cause": "memory leak in payment-service"}' # 4. Fix it (reward: +0.40) curl -X POST https://huggingface.co/space_proxy/arijit-07-devops-incident-response/step -H "Content-Type: application/json" -d '{"action_type": "restart_service", "service": "payment-service"}' # Score: ~0.94 ✅

import requests BASE = "https://arijit-07-devops-incident-response.hf.space" # Start episode obs = requests.post(f"{BASE}/reset", json={"task_id": "easy", "seed": 42}).json() # Take action result = requests.post(f"{BASE}/step", json={"action_type": "read_logs", "service": "payment-service"}).json() print(f"Reward: {result['reward']}") # 0.15

🧠 Training Evidence

Before vs After

Base Llama-3.1-8B0.000

❌

jumps to diagnose, gets penalized

ARIA Fine-tuned0.150

✅

reads logs first, every time

Model weights →

Training Details

Algorithm

GRPO

Base Model

Llama-3.1-8B-Instruct

Framework

Unsloth + HuggingFace TRL

LoRA Rank

32 (alpha 64)

Episodes

160

GPU

NVIDIA L4