2akouwu/reverify

What it solves

Reverify addresses the problem of AI hallucinations in reverse engineering (RE). When language models are asked to reconstruct algorithms or structures from a binary, they often invent offsets, sizes, and behaviors. Reverify ensures that AI-generated hypotheses about a binary are grounded in actual bytes rather than the model's imagination.

How it works

The project pairs a language model with a deterministic, pure-Python reverse engineering toolkit. The AI proposes a hypothesis (a "claim"), and the toolkit acts as the judge, checking the claim against the actual binary data. A claim is only reported as verified if the tools—which handle disassembly, pattern matching, or CPU emulation—confirm it.

It can be used as a CLI tool or as an MCP server, allowing AI agents like Claude Code or Cursor to call these tools directly to verify their own findings.

Who it’s for

Security researchers, malware analysts, and CTF participants who need to use AI to assist in reverse engineering without sacrificing accuracy or trusting hallucinated data.

Highlights

• Verification Loop: A closed-loop system where a model proposes claims and the tools iterate until the findings are grounded in the binary.
• Deterministic Core: Built-in support for PE/ELF/Mach-O parsing, x86/x64/ARM/ARM64 disassembly, and CPU emulation.
• Agent-Native: Ships as an MCP server for direct integration with AI agents.
• Flexible Backends: Uses a pure-Python core by default, but can be upgraded to use Capstone, Unicorn, and LIEF for more mature engine support.