Md Ismail Šojal (@0x0SojalSec) introduces a locally runnable Cyber-OSINT model on X. Built by Deep Hat, this 26B parameter model (4B active) is trained on 6,500 OSINT and CT instructions. It supports threat-actor attribution, IOC pivoting, geolocation, and Admiralty source grading with a 262K context window.

A separate 7B cyber model fits on an 8GB GPU and is fine-tuned for offensive and defensive DevOps tasks. It features 32K native context, expandable to 131K with YaRN. Unlike many security models that rely on system prompts, this is a genuine fine-tune.

Available on Hugging Face at huggingface.co/Deep Hat/Deep Hat.