Noelo-Lab/kuna

Kuna – An Agent‑First Decompiler (Rust)

What it is

• Kuna is a binary decompiler written in Rust. It takes compiled executables (e.g., ELF, PE) and produces C source, header, and assembly files.
• The project is built around the idea that large‑language‑model (LLM) agents will be the primary users of decompilers, so the tool is designed to be agent‑first: fast, text‑oriented, and easily scriptable.

Key features (as described in the README)

• Speed & accuracy – Holds the top spot on the DecBench benchmark for optimized C and can decompile a 19 MB binary in under three minutes.
• Multiple front‑ends – Distributed as a single self‑contained Rust binary that can be run:
  1. CLI – kuna decompile <binary> <function> with optional flags (--addr, --option).
  2. Web (WASM) – A browser‑based UI at kuna.noelo.org/decompile that runs entirely client‑side, keeping binaries private.
  3. Ghidra plugin – Can replace Ghidra’s built‑in decompiler via the KunaDecompiler extension.
• Agent integration – Ships a “skill” file (skills/kuna/SKILL.md) that LLM agents like Claude Code, Codex, or OpenCode can install to learn the optimal command usage.
• Extensible pipeline – The decompiler is split into clearly defined phases (p0_knowledge … p9_emit). Each phase is documented in natural‑language specs, making it easier for LLMs to propose and test changes.
• Tunable options – Feature toggles are exposed through a hierarchical option catalog (docs/options.md). This lets different users (e.g., reverse‑engineers vs. exploit developers) request high‑level or low‑level output.

How to get it

• Pre‑built binaries for Linux, Windows, and macOS are available on the GitHub Releases page.
• Building from source requires only a Rust toolchain (cargo). The repository provides a Makefile with shortcuts:
  • make binaries – builds the decompiler, SLEIGH compiler, and CLI.
  • make specs – compiles the SLEIGH processor specifications.
  • make test – runs the full regression suite (675 assertions across 83 test files).

Project goals

1. Autonomous refinement – Use LLM‑driven pipelines and DecBench metrics to continuously improve the decompiler without human intervention.
2. Agent‑first design – Prioritize decompilation text quality and speed, providing a clean interface for LLM agents.
3. Tunable pipelines – Allow agents (or humans) to enable/disable features on the fly to suit different reverse‑engineering tasks.

Community & support

• Active Discord channel for questions and development discussion.
• Sponsored by the Air Force Research Lab, Metalware, and the U.S. Department of Defense.
• Contribution guidelines and agent‑facing documentation are in AGENTS.md and docs/.

Bottom line Kuna is a modern, Rust‑based decompiler that positions itself as a backend for AI agents doing reverse engineering. It offers fast, high‑quality C output, multiple deployment options, and an architecture built for automated, LLM‑driven improvement.