Salt: Systems programming language with Z3 theorem proving in the compiler.

What it does

Salt is a systems programming language that embeds Z3 theorem proving into its compiler. The language lets developers write preconditions and postconditions in code, then verifies them at compile time rather than runtime. It targets C-level performance and supports arena memory allocation and MLIR-based code generation to native binaries.

Who it is for

Salt targets systems programmers building performance-critical software where memory safety and bounds checking matter. The language is positioned for developers who want formal verification guarantees without garbage collection or borrow checkers. The creator has demonstrated it on low-level work: a Llama 2 inference engine, a microkernel for x86 hardware with SMP scheduling, and a neural network trainer.

Pricing

The site does not show prices.

How it stands out

Most systems languages choose between safety and speed. Rust uses a borrow checker; C offers neither safety nor compile-time verification. Salt claims to offer mathematical proof of correctness for array bounds and other invariants at compile time, verified by Z3, without runtime overhead. The demonstration projects show concrete results: a 600-line Llama 2 implementation with verified kernels, a 16-core microkernel running on real x86 hardware with measured context-switch times around 487 cycles, and a 188-line neural network trainer achieving 97% accuracy on MNIST.

The language is open source on GitHub and includes a live editor. The creator built KeuOS, a real operating system kernel written in Salt, to validate the compiler's guarantees in practice.

What a founder should check

A competitor would need to verify three things. First, whether Z3 theorem proving actually scales to large codebases without creating compile-time bottlenecks; the demonstrations are relatively small. Second, what the actual adoption friction is: do developers prefer Rust's familiar borrow-checking model despite its learning curve, or will they adopt Salt's proof-based approach? Third, whether the performance claims hold across a broader range of workloads beyond the specific benchmarks shown (inference, kernels, training). The gap between a 600-line kernel demo and production systems code is substantial.

Thinking of building something like this?

Every launch here is a competitor to somebody's idea. If yours is close, check it against the market before you build: the Full Check names the rivals, the prices and the gaps.