In the post-AGI economy, humans will eventually stop writing and reading code, but we still need an ambiguity-free language to communicate our intents to the AIs building the world around us. Bend is that language.

With laws, intents can be more precise than natural language. With proofs, we can mechanically verify the AI implemented our prompts correctly. And with a fast compiler, we can run that code at peak compute.

That's Bend - and nothing else.

Bend runs FAST

Target: be as fast as C on the CPU, as fast as CUDA on the GPU. Status:

Thanks to strong types, purity and linearity, Bend compiles to fast executables as fast as hand-written C (single-core), and even faster (on 10000s cores). The entire language runs on the GPU, with full memory unification.

Bend checks FAST

Target: outperform every proof assistant by several OOMs. Status:

Bend's compiler is so powerful it can verify mathematical proofs. Usually, this is slow. Bend is not. It checks, in under a second, files that other projects would take minutes, making proofs way more practical.

Bend is PARALLEL

No threads, no locks, no kernels to write. Split the work in two, and Bend spreads the calls over every core it can find, then joins them back. Below, pow2(20) divides until one task sits on each of 4,096 GPU cores:

Bend BLOCKS mistakes - with proof

PROBLEM: How can you trust AI code, without reading it?

SOLUTION: By forcing your AI to write a correctness proof.

Bend introduces LAWS.bend, a file where you declare rules that your app must not break. Bend's compiler then guarantees that these laws always hold, by demanding mathematical proof whenever your code is edited. For example, consider a game with one law: winning is impossible. Here's how it plays out:

Without LAWS.bend, a bug was merged. With it, the AI had to retry, until no bugs were left! In this case, it added a wall, but it could have moved the flag, made the room kill you, or whatever. The only thing it can't do is commit a bug, because it is mathematically impossible to break laws in LAWS.bend. The compiler enforces it.

Using LAWS.bend is simple.

1. Ask your AI to formalize your app's rules in LAWS.bend. Example:

   • LAW: "the sum of all balances must be zero"

   • LAW: "players can never pass through solid walls"

   • LAW: "list_sort() must always return ascending numbers"

   • LAW: "array_set() may never be called out-of-bounds"

   • LAW: "winning is impossible" (the demo above!)

   • And so on. Anything you can spell can become a law.

2. Ask your AI to run bend PROOF.bend after editing any code.

3. That's it. Rejoice as your app never again breaks or violates your rules.

You can also edit LAWS.bend yourself. Here's how it looks:

# LAWS.bend
law you_cant_win:                           # "winning is impossible"
  for moves: List<Game.Move>                # any sequence of moves
  board = Game.replay(Game.start(), moves)  # replayed from the start
  {Game.is_won(board) == False{} : Bool}    # never leads to victory

# PROOF.bend
def Laws.you_cant_win(moves):
  # ... written by the AI

In short, LAWS.bend is AGENTS.md backed by proof.

With LAWS.bend, "make no mistakes" becomes enforceable.

Skeptical? Edit the demo's code and break the "you can't win" law!

Get Started

1. Install:

curl -fsSL https://bend-lang.com/install.sh | sh

This bend ignores a project's bunfig.toml and .env; bun bend2/main.ts from a checkout reads them, so check untrusted code with bend.

2. Tell your agent to use Bend:

Add this to your AGENTS.md:

When using Bend:
- run `bend guide` to learn it
- use `LAWS.bend` to keep important rules
- run `bend PROOF.bend` before committing
- parallelize the code whenever possible

Then, just say: "use Bend"!

3. Enjoy bug-free, fast vibe-coded apps!

Hints:

• Ask it to write laws for anything that can't break.

• Ask it to parallelize anything you want to be fast.

• Bend is young. If anything goes wrong, ask it to open an issue. <3

Bend works best on the back-end, on Linux or macOS.

Examples

Syntax == Python + dependent types

import Base

# Performs effects on the CPU.
def main() -> IO(Unit):
  do IO<Unit>:
    name : String <- IO.try(String, IO.get_env("USER"))
    IO.print("Hello, " ++ name)

Parallelism == divide-and-conquer

import Base

# Computes 2^d in parallel: a tree of d levels, one leaf per unit.
def pow2(+d: Nat) -> U32:
  match d:
    case 0n:
      1
    case 1n+p:
      a b = pow2(p) pow2(p)
      (a + b : U32)

# Runs pow2 on the GPU, via `!`.
def main() -> IO(Unit):
  result = pow2!(20n)
  IO.print(U32.show(result))

Theorems == laws, Proofs == defs

import Base

# CLAIM: for every nat x, x + 0 equals x.
law add_zero:
  for x: Nat
  {Nat.add(x, 0n) == x : Nat}

# PROOF: induction on `x`, one rewrite (`%`) per step.
def add_zero(x):
  match x:
    case 0n:
      {==}
    case 1n+xp:
      %add_zero(xp) : {1n+Nat.add(xp, 0n) == 1n+_ : Nat}
      {==}

This is the beginning of the README. Read the rest at the original. Read the full README ↗