You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

ROLLY-AGENTIC-CYBER-DATASET

Curator: ROVALTHIA LABORATORY
Official Dataset Repository: JuanCDEV/rovalthia-data
Companion Model: JuanCDEV/rovalthia
License: Apache 2.0
Domains: Defensive Cybersecurity, Zero Trust Architecture, CTI Threat Hunting, Autonomous Agentic Orchestration, Formal Reasoning.


1. Overview & Purpose

The ROLLY-AGENTIC-CYBER-DATASET provides a vetted, high-density instruction and evaluation dataset designed to train and benchmark autonomous agentic AI models on defensive security, incident response, vulnerability management, and multi-agent coordination.


2. Splits & Composition

Split File Examples Count Purpose
Train data/train.jsonl 244 Supervised instruction tuning for defensive cyber reasoning
Validation data/validation.jsonl 52 In-training loss and early stopping evaluation
Test data/test.jsonl 35 Public diagnostic benchmark split with ground-truth rubrics

All private enterprise documents and proprietary customer logs are strictly excluded.


3. Provenance & License Matrix

Every sample in this dataset is auditable and traceable to permissible open-source or public domain research:

  • SANS Technology Institute Academic Research (Permissive citation)
  • NIST Special Publications 800-Series (United States Public Domain)
  • CISA Advisories & KEV Catalog (United States Public Domain)
  • Synthetic Multi-Turn Reasoning Traces (ROVALTHIA Proprietary, released under Apache-2.0)

Full license audit: LICENSE_MATRIX.json.


4. How to Load

from datasets import load_dataset

dataset = load_dataset("JuanCDEV/rovalthia-data")
print("Train split size:", len(dataset["train"]))
print("Sample:", dataset["train"][0])
Downloads last month
14