AILang

An AI-native systems programming language — built for AI to understand, generate, and maintain.

🌐 English | 简体中文

Pre-Alpha · design phase　Spec v0.2.1 (frozen)　Reference compiler: not yet

---

What it is

AILang is a systems programming language for the AI era. It compiles to native binaries via LLVM — no VM, no hard GC — targeting Rust-level safety and performance.

Its core idea: code should actively express intent. Types carry meaning, functions declare their effects, errors come from the signature, ownership is explicit — and the compiler turns all of it into machine-readable metadata (.ailmeta) that AI agents can trust like a type signature.

AILang = Python's simplicity + C/Go braces + Rust's safety + TypeScript's types + AI-native semantics.

---

At a glance

package app

type UserId = int
meaning UserId:
    "unique identifier of a user"

struct User {
    id: UserId
    name: string
}

error UserError {
    NotFound
    PermissionDenied
}

fn get_user(id: UserId) -> Result<User, UserError>
effects {
    database.read
}
{
    return database.find(id)
}

Compiling this produces a native binary and an app.ailmeta:

{
  "name": "get_user",
  "input":  [{ "name": "id", "type": "UserId", "meaning": "unique identifier of a user" }],
  "output": "Result<User, UserError>",
  "effects": ["database.read"],
  "errors": ["NotFound", "PermissionDenied"]
}

An AI reads this structured, compiler-verified description instead of guessing from source. That is the whole reason AILang exists.

---

Key features

• 🤖 AI-native types — semantic types (type UserId = int + meaning), constraint types, an effect system, errors derived from return types.
• 📄 .ailmeta — semantic facts verified by the compiler (not comments). AI trusts them like a type.
• 🛡️ Rust-level safety, no lifetimes — ownership & borrow with inferred lifetimes; borrows don't escape, so there's no 'a.
• ⚡ Rust-level performance — LLVM backend, native binaries, zero-cost abstraction.
• 🐍 Python-level simplicity — braces {} + optional semicolons.
• 🧩 AI as a first-class citizen — agent / tool declarations auto-generate OpenAI Tool Schema.

---

Status

In design phase. The spec is frozen (v0.2.1), but there is no reference compiler yet — you can't ail build today. Spec review and design feedback are very welcome; compiler work (Phase 1) starts soon.

---

Documentation

The complete design lives in a single document — docs/AILANG.md (v0.2.1, frozen). It's organized in six parts:

• Part 0 · Motivation & Philosophy
• Part I · Language Specification — the authoritative spec (incl. ownership & concurrency deep-dives)
• Part II · Standard Library & Packages
• Part III · Tutorial (runnable examples)
• Part IV · Compiler Implementation Design
• Part V · Decision Records

First time here? Start with Part 0, then read Part I.

---

License & contact

Apache License 2.0 — see LICENSE.

• Repository: https://github.com/miuiadmin/ailang
• Discussions: https://github.com/miuiadmin/ailang/discussions

---

AILang believes code should express intent, not just instructions — and that performance, safety, and AI-legibility need not be traded against each other.