The Liya-IL Manifesto

Liya‑IL

Position, Not Syntax.
The First AI-Native Language.

Scroll Down
— Manifesto

Code Is No Longer Written by Humans

In 2026, AI writes the code. Humans only state intent. Liya-IL is designed for this new reality.

Position Beats Syntax

def, function, async, await, try...except words are now unnecessary. In Liya-IL, a symbol's meaning comes from where it stands . Up, in, down, beside — position determines meaning.

Two Layers, One Language

Deep Layer (L1): Raw IDs. One token for AI-to-AI communication. Surface Layer (L2): Human-readable text, emoji, or braille. Same meaning, different form.

Deep Layer

AI-to-AI

🔁📋   🔍📋⚖️🔢   📡📋

Fixed 2-byte tokens. Position-based grammar. Zero ambiguity, maximum efficiency.

Surface Layer

Human View

fn reverse_evens(arr: [int]) -> [int] {
  arr.rev().filter(|x| x % 2 == 0)
}

Text, emoji, SVG, or braille. Customizable per user preference. Same L1 token underneath.

🗣️
Human Intent
🌐
LiyaAI Gateway
🔁📋
Liya-IL Iconic
🧠
Liya-Core
📤
Target Code
Python~14 tokens
def reverse_and_get_evens(arr):
    reversed_arr = list(reversed(arr))
    return [x for x in reversed_arr
            if x % 2 == 0]
Liya-IL Text~7 tokens
fn reverse_evens(arr: [int]) -> [int] {
  arr.rev().filter(|x| x % 2 == 0)
}
Liya-IL Iconic1 token
🔁(🔍(📋, ⚖️🔢))
14Python Token
7Liya-IL Text
1Liya-IL Iconic
14×fewer tokens

Accessibility Is Not a Privilege, It Is a Fundamental Right

Blind developers are first-class users. STT/TTS natural flow, screen reader optimization, and voice code generation are integrated from day one.

— manifesto.principles.section_label

manifesto.principles.heading

🤖

AI-Producible

Position-based grammar means zero syntactic ambiguity. AI models produce correct code on the first try.

🧑‍💻

Human-Debuggable

When things go wrong, a human can read and fix it. Clean syntax that never sacrifices clarity.

Token-Efficient

3–5× fewer tokens than Python for equivalent logic. Faster generation, lower cost.

🛡️

Type-Safe

The type system catches errors before runtime. Immediate feedback during AI generation.

🎙️

STT/TTS Native

Designed for voice-first workflows. Blind developers are first-class users.

🌍

Cross-Compilation

Compiles to every target architecture via LLVM backend. From WebAssembly to ARM — one language, everywhere.

Language is the shape of thought. Redesigning the language AI thinks in is redesigning the future.
— Liya-IL Manifesto, 2026
— Roadmap

The Path Ahead

Phase 0

Manifesto & Specification

Manifesto, language grammar (BNF/EBNF), GitHub repo, and RFC process publication.

Phase 1

Dataset Generation

10,000+ natural language → Liya-IL → target code examples with unit-test validation.

Phase 2

Minimal Interpreter

Python-based lexer + parser + AST. Interactive playground at il.liyalabs.com.

Phase 3

Liya-Core Fine-tuning

QLoRA fine-tuning on Qwen 3.5 Coder 7B. vLLM deployment for sub-millisecond inference.

Phase 4

Open Source Community

100+ contributors. Standard library, multilingual gateway expansion.

Let's Build the Future Together

Liya-IL is open source and community-driven. Contribute, share your ideas, and shape the evolution of the language.