REAgent — AI-Powered Reverse Engineering Assistant
591 hand-crafted training examples across 20+ specialized security domains, built on a custom fine-tuned 27B parameter model.
What it knows:
Reverse Engineering — PE analysis, disassembly, decompilation, string decryption, pattern scanning,
cross-references, IDA Pro & Ghidra workflows, binary diffing
Unpacking & Deobfuscation — UPX, Themida, VMProtect, custom packers, control flow flattening, VM
devirtualization, aPLib/LZMA decompression, multi-layer unpacking
Malware Analysis — Dropper/RAT/ransomware/rootkit identification, shellcode analysis, API hashing, config
extraction, sandbox evasion, persistence mechanisms
Windows Kernel Internals — EPROCESS/ETHREAD structures, IOCTL dispatch, minifilters, PatchGuard, DSE
bypass, SSDT/IDT hooks, kernel callbacks, physical memory access, page table walking
Anti-Cheat Analysis — Usermode & kernel-mode anti-cheat architecture, detection methods (integrity checks,
timing, heartbeats, module scans), bypass techniques, callback manipulation
Hypervisor & Virtualization — Intel VT-x/EPT hooking, VMCALL dispatch, EFI boot chain hooks, memory hiding,
CR3 manipulation, PatchGuard disable — trained on real hypervisor source code
Game Security Research — Entity lists, pointer chains, signature scanning, struct recovery, HWID spoofing,
DMA/FPGA analysis, overlay rendering
Hooking & Injection — IAT/VMT/trampoline hooks, direct syscalls, process/DLL hollowing, thread hijacking,
manual mapping, instrumentation callbacks
Cryptography — AES/RC4/XOR identification, CryptoAPI/BCrypt usage, certificate pinning, protocol reversing,
encrypted config extraction
Agent Discipline — Knows when to use which tool, handles errors gracefully, doesn't hallucinate findings,
writes compact scripts, asks for help when stuck
---
Not a generic coding model with security knowledge bolted on. Every example was built specifically for RE
workflows — from initial binary triage to kernel-level exploitation analysis.
Price model will be based on per token used.
Out here to scope if there is potential for this or if it's just for myself, you still need basic knowledge and understanding to direct the model to do what you want to do.
IM ALSO LOOKING FOR PEOPLE WITH KNOWLEDGE AND TRAINING EXAMPLES TO HELP DEVELOP THIS EVEN MORE.
Best examples are working examples. Compensation provided, can discuss details in PM. Only interested in modern/recent examples, old methods dont really matter.
 |
