AI programming language
A16
A performance-first language for fast, stable AI.
A programming language and runtime designed for AI workloads where execution speed, stability, and resource efficiency are central requirements.
Proprietary softwareProject overview
What A16 does.
A16 combines a Python-like language with an optimized bytecode virtual machine and a modular Rust implementation. Its repository documents built-in tensors, automatic differentiation, embeddings, vector memory, an agent runtime, async execution, native extensions, and IDE tooling.
The project is proprietary software. Its repository states that no permission is granted to use, copy, modify, or distribute it without written permission from the copyright holder.
Verified capabilities
Explore the product.
01AI-native primitives+
Built-in N-dimensional tensors, autograd, embeddings, vector memory, and agent-oriented language features.
02Language and runtime+
A lexer, parser, type checker, intermediate representation, bytecode compiler, and stack-based virtual machine.
03Developer tooling+
Command-line tools for running, checking, formatting, linting, documenting, parsing, and inspecting A16 programs, plus an interactive REPL and language server.
04Concurrency and extensions+
Async execution, channels, a package system, and foreign-function interfaces for native extensions.
05Self-hosting work+
The project includes a bootstrap compiler written in A16 with lexer, parser, code generation, and virtual-machine components.
Language example
A16 in a few lines.
This tensor example is published in the A16 repository.
fn main():
let x = tensor_ones([2, 3])
let w = tensor_randn([3, 2])
let y = tensor_matmul(x, w)
let out = tensor_relu(y)
println(tensor_shape(out))
return 0