Back to Portfolio
BinaryVibes
experimentalAIPythonAssemblyCLILLM

BinaryVibes

From natural language to native binary — skipping the high-level compiler.

No high-level compiler — natural language → assembly → native binary

terminal
$ bv build "fetch weather for Seattle, London, and Tokyo and print each"
# → Seattle: 🌦 +4°C | London: ☁️ +6°C | Tokyo: ⛅️ +8°C
# That is a 4KB native .exe — no compiler involved.

$ bv build "hello world" --format elf -O hello # Linux
$ bv build "hello world" --format macho -O hello # macOS
$ bv build "hello world" --format pe -O hello.exe # Windows

An LLM-driven synthesis framework that generates native executables from English — no C, Rust, or Go compiler required. The pipeline goes: natural language → LLM → x86_64 assembly → Keystone assembler → PE/ELF/Mach-O binary, with a two-level self-correcting feedback loop for assembly errors and runtime crashes.

How it works

The interesting claim isn't 'no toolchain at all' — it's that an LLM can reliably generate correct, working x86_64 assembly for non-trivial programs (HTTP networking, GUI dialogs, file I/O) with a self-correcting feedback loop. What's skipped is the high-level language compiler — no GCC, Clang, or MSVC. What still runs: Keystone assembles the LLM output into machine code bytes, and a custom PE/ELF/Mach-O builder handles headers, import tables, and section layout. 14 pre-baked helper routines solve the hardest part: register preservation and stack alignment across Windows API calls. The result is a 2–4KB native executable from a plain English description, with no human-written source code in between.