jordanhubbard/nanolang

A tiny experimental language designed to be targeted by coding LLMs

What it solves

NanoLang is a programming language and secure runtime designed specifically to be written by machines (such as LLMs) and read by humans. It addresses the problem of ambiguity in programming languages, which can lead to errors when AI agents generate code. It also provides a secure, least-privilege environment for hosting services on standard kernels without requiring a custom kernel.

How it works

NanoLang uses an unambiguous syntax and a core semantics that are formally proved in Coq (covering type soundness, progress, and determinism). It can be transpiled to C for native performance or executed via the NanoISA virtual machine, a stack-based VM that isolates dangerous external calls in a separate process to ensure stability.

Who it’s for

It is intended for developers who need a formally verified, secure language for machine-generated code, as well as those interested in formal verification, secure runtimes, and the intersection of AI-driven development.

Highlights

  • Formally Proved Semantics: Core semantics are mechanically proved in Coq with no axioms or admitted sub-cases.
  • NanoISA VM: A verified bytecode VM that isolates FFI calls in a co-process to prevent crashes from affecting the main program.
  • Machine-Led Optimization: Supports constant folding, dead-code elimination, and profile-guided inlining.
  • Secure Runtime: Features unforgeable capabilities, a POSIX service fabric, and a trap journal for secure service hosting.
  • Shadow Tests: Mandatory test blocks that accompany function definitions to ensure code is tested by design.
  • LLM-Ready: Includes a machine-readable formal specification (spec.json) and a training reference (MEMORY.md) to help LLMs write the language correctly.

Related

  • Dispatch
    nanoVLM: A Minimalist PyTorch Toolkit for Training Vision Language Models

    Hugging Face has released nanoVLM, a lightweight, pure PyTorch toolkit designed to simplify the training and understanding of Vision Language Models (VLMs) for beginners and researchers.

  • Project
    QwenAudio/CosyVoice

    CosyVoice is an advanced, LLM-based text-to-speech system designed for high-quality, zero-shot multilingual speech synthesis and voice cloning.

  • Project
    he-yufeng/CoreCoder

    A minimal, 1.2k-line Python engine that implements a coding agent, designed as a readable reference for developers to learn how agents work and fork their own.

  • Project
    gakonst/nanocodex

    Nanocodex is a library‑first SDK (Rust core with JS/Node, browser WASM, and Python bindings) that implements the complete OpenAI coding‑agent loop—session‑wide typed history, streaming events, tool integration, branching, snapshots, and optional durability—so applications can embed a fully‑featured LLM coding assistant without re‑creating the orchestration plumbing.

  • Project
    OpenRaiser/NanoResearch

    An end-to-end autonomous AI research engine that automates the process from ideation to LaTeX paper production by executing real GPU experiments to ensure data authenticity.