NeverD
A binary analysis and decompilation engine built on LLVM that lifts native and smart-contract code to one-to-one instruction-level semantics, for use by command-line tools, other programs and AI agents through a single C API.
NeverD loads PE, ELF, Mach-O, legacy EVM bytecode and Solana SBF ELF programs, decoding native targets with Capstone and using dedicated version-aware decoders for EVM and SBF, then represents every instruction with hand-written semantics rather than an approximation. Supported instructions can be rewritten into LLVM IR, C, Rust for SBF, Solidity-oriented EVM reconstruction, or, for native targets, a rewritten binary; strict mode is on by default, so an instruction with no lifter raises an error instead of being skipped or guessed.
Callers, including AI agents, use one engine, libneverd, through a pure C API without linking Capstone, LLVM or internal C++ directly. It also includes an experimental Android recovery mode that exports Java sources from APK and DEX inputs, and an experimental iOS workflow that exports native C and Objective-C/Swift sources from an IPA or Mach-O binary.
It is written in C++20.