LLMLang is a new programming language designed from the ground up to be optimally friendly for Large Language Models (LLMs) and code generation tools.
This repository presents the founding doctrine: every contributor and ecosystem maintainer must adhere to these principles when developing the language, standard library, tools, or any infrastructure.
- Every syntax element must have only one clear, explicit interpretation.
- For each task, there must be a single canonical way to express it—no syntactic "magic" or shortcuts.
- All operations must be explicit; implicit behaviors and hidden conversions are forbidden.
- All constructs must follow a uniform style (e.g.,
call function args to resultfor all function/method calls). - New features must conform to the established patterns of the core language and standard library.
- Behavior must be consistent regardless of platform or context.
- All variables, struct fields, function results, and array values must be explicitly named and declared.
- Assignments always use a clear operator (such as
to), never overloaded or ambiguous operators like=. - No implicit global or hidden state; all data must be passed explicitly via parameters or named variables.
- Every language construct must be easily parsable by LLMs and automated tools.
- No overloaded operators or multi-purpose symbols.
- Any action should be expressible as a simple, single-purpose command or function.
- Only simple, canonical control structures (
if,loop,call,foreach, etc.). - All nesting must be explicit and stepwise; avoid complex nested expressions or closures.
- No superfluous symbols or sugar: every character in the syntax serves a clear purpose.
- Syntax is chosen for machine clarity, not for human aesthetics.
- All new functions, structures, and modules must be specified according to the same rules as the language core.
- Extensions are always implemented as separate modules/files—never as hidden or magical features.
- All documentation and specifications must be structured to be as easily parseable by LLMs as by humans.
- For every language and library element, inputs, outputs, side effects, and constraints must be explicitly described.
- The language must not allow implicit state changes or side effects without explicit notation.
- No hidden contexts, closures, or non-obvious references.
- All improvements or changes to the language, compiler, libraries, or tooling must be evaluated first from the perspective of LLMs and automated code generation, not just humans.
- Priority: minimize error risk for code generation, maximize ease of code analysis and auto-completion, ensure compatibility with static analysis and automated testing.
LLMLang is a language where every operation has one, absolutely transparent and predictable form, and every part of the language is designed primarily for machines—not humans.
All contributors must follow this doctrine and must not introduce anything that reduces the formality, unambiguity, or predictability of the language for LLMs and automated tools.
If you want to join the development or suggest enhancements, please open an issue or a pull request.
Every proposed change must explicitly state how it aligns with the LLMLang doctrine.