Joule

A self-hosted programming language with compile-time energy budgets.

Joule is the world's first energy-aware programming language with first-class compile-time energy budget enforcement. Every function has an energy cost. Every accelerator is measured. Every decision is accountable. The Joule self-host bootstraps from a small kernel and lowers through HIR to native (C / Cranelift / LLVM / MLIR / CUDA), WASM, and flowG.

This is the public release surface. Source is private at openIE-dev/joule-lang-core — a 161-crate nested workspace including the self-host bootstrap, codegen passes, runtime, edge tooling, and the joule cascade.

Status

Release binaries + examples + documentation. Joule is free to use software, not an open-source project. See LICENSE for Business Source License 1.1 terms — converts to Apache-2.0 four years after each binary's release date.

What you get

Plus a Rust library, joulec, for embedding.

Install

# via cargo
cargo install joulec

# via cargo-binstall (prebuilt binary)
cargo binstall joulec

# direct download
curl -fsSL https://github.com/openIE-dev/joule-lang/releases/latest/download/joulec-$(uname -s)-$(uname -m).tar.gz | tar xz

Platform support:

Hello, Joule

fn add_one(x: f64) -> f64
    @energy(< 100pJ)
:
    x + 1.0

fn main():
    let r = add_one(41.0)
    println(r)

joulec build examples/hello/add_one.joule
./add_one

Energy receipts:

joulec build examples/hello/add_one.joule --record-energy receipt.json
cat receipt.json | jq '.functions[] | {name, budget_pj, measured_pj}'

See examples/ for hello world, multi-backend codegen, the bootstrap walkthrough, and the joule → flowG bridge.

What makes Joule different

• Compile-time energy budgets — @energy(< 100pJ) annotations are enforced by the compiler against measured per-op cost data. Budget violations fail to build.
• Self-hosted bootstrap — the Joule compiler is written in Joule. The bootstrap C emitter starts from a small kernel and grows out. You can audit the self-host trace.
• Multi-backend codegen — same .joule file can lower to C, Cranelift, LLVM, MLIR, CUDA, WebAssembly, or flowG. Choose your target by use case; the energy budget tracks across backends.
• Joule cascade — at runtime, Joule programs participate in a 16-level cascade (L0–L10) that escalates from cache → lawful synthesizer → small models → frontier as needed.
• Edge-ready — Joule binaries are small enough to ship to drones, microcontrollers, OpenMV cameras, and other embedded targets.

How it fits

Joule is the energy-budgeted compiled surface in the openIE-dev family:

• flowG — the substrate Joule lowers to
• Lux — reactive, app-oriented sibling
• JMax — math-native sibling
• JouleDB — the metered database that Joule programs persist to

Documentation

• Getting started → https://openie-dev.github.io/joule-lang
• Examples → examples/
• Language reference → https://openie-dev.github.io/joule-lang/reference
• Bootstrap walkthrough → https://openie-dev.github.io/joule-lang/bootstrap
• Source mirror → openIE-dev/joule-lang-core

Releases

GitHub Releases — tagged versions with prebuilt binaries for every supported platform, plus SHA-256 checksums.

Community

• Discussions — Q&A, language design
• Issues — bug reports
• Security: see SECURITY.md

License

• Binaries — Business Source License 1.1; see LICENSE. Free for non-commercial use, internal use by orgs under $1M revenue, security/academic research. Converts to Apache-2.0 four years after each release.
• Documentation — CC-BY-4.0
• Examples — Apache-2.0