AI-optimized systems programming language with token-efficient syntax.

Vais is designed to minimize token usage while maximizing code expressiveness, making it ideal for AI-assisted development and LLM code generation.

• Single-letter keywords - F (function), S (struct), E (enum/else), I (if), L (loop), M (match)
• Self-recursion operator @ - Call the current function recursively
• Expression-oriented - Everything is an expression
• LLVM backend - Native performance with LLVM 17
• Type inference - Minimal type annotations with full constraint solving
• Memory Safety - Borrow checker with Non-Lexical Lifetimes (NLL), --strict-borrow mode
• Slice Types - &[T] / &mut [T] with fat pointer implementation
• Parallel Compilation - DAG-based parallel type-check and codegen (2-4x speedup)
• Self-Hosting - 50,000+ LOC bootstrap compiler, 21/21 clang 100% success
• Rich Ecosystem - 29 crates, 74 stdlib modules, growing package ecosystem

# Fibonacci with self-recursion
F fib(n:i64)->i64 = n<2 ? n : @(n-1) + @(n-2)

# Struct definition
S Point { x:f64, y:f64 }

# Sum with loop
F sum(arr:[i64])->i64 {
    s := 0
    L x:arr { s += x }
    s
}

crates/
├── vais-ast/          # AST definitions
├── vais-lexer/        # Tokenizer (logos-based)
├── vais-parser/       # Recursive descent parser
├── vais-types/        # Type checker & inference
├── vais-codegen/      # LLVM IR code generator (inkwell/, advanced_opt/)
├── vais-codegen-js/   # JavaScript (ESM) code generator
├── vais-mir/          # Middle IR
├── vaisc/             # Main compiler CLI & REPL
├── vais-lsp/          # Language Server Protocol
├── vais-dap/          # Debug Adapter Protocol
├── vais-jit/          # Cranelift JIT compiler
├── vais-gc/           # Optional garbage collector
├── vais-gpu/          # GPU codegen (CUDA/Metal/OpenCL/WebGPU)
├── vais-i18n/         # Internationalized error messages
├── vais-plugin/       # Plugin system
├── vais-macro/        # Declarative macro system
├── vais-hotreload/    # Hot reloading
├── vais-dynload/      # Dynamic module loading & WASM sandbox
├── vais-bindgen/      # FFI binding generator (C/WASM-JS)
├── vais-query/        # Salsa-style query database
├── vais-profiler/     # Compiler profiler
├── vais-security/     # Security analysis & audit
├── vais-supply-chain/ # SBOM & dependency audit
├── vais-testgen/      # Property-based test generation
├── vais-tutorial/     # Interactive tutorials
├── vais-registry-server/    # Package registry (Axum/SQLite)
├── vais-playground-server/  # Web playground backend
├── vais-python/       # Python bindings (PyO3)
└── vais-node/         # Node.js bindings (NAPI)

std/               # Standard library (79 modules)
selfhost/          # Self-hosting compiler (50,000+ LOC)
vscode-vais/       # VSCode extension
intellij-vais/     # IntelliJ plugin
docs-site/         # mdBook documentation
examples/          # Example programs (174 .vais files)
benches/           # Benchmark suite (criterion + language comparison)
playground/        # Web playground frontend

cargo build --release
cargo test                                     # Run all 9,300+ tests
cargo test -p vaisc                            # Run vaisc tests (1,620+ E2E tests)
cargo clippy --workspace --exclude vais-python --exclude vais-node

This project uses cargo-llvm-cov to measure test coverage. Coverage reports are generated automatically in the CI pipeline.

To generate coverage reports locally:

# Install cargo-tarpaulin (one-time setup)
cargo install cargo-tarpaulin

# Generate coverage reports (HTML and Lcov)
cargo tarpaulin --config tarpaulin.toml

# Or use the convenience alias
cargo coverage

# Generate HTML report only
cargo coverage-html

# Generate Lcov format for CI integration
cargo coverage-lcov

Coverage reports are saved to target/coverage/:

• index.html - Interactive HTML coverage report
• lcov.info - Lcov format for codecov integration

Coverage is measured automatically on every push and pull request to main and develop branches. Reports are:

• Uploaded as GitHub Actions artifacts
• Sent to Codecov for tracking trends
• Available for 30 days in the CI artifacts

# Compile a Vais file
./target/release/vaisc build hello.vais -o hello

# Run directly
./target/release/vaisc run hello.vais

# Start REPL
./target/release/vaisc repl

# Format code
./target/release/vaisc fmt src/

# Check for errors
./target/release/vaisc check hello.vais

• Lexer (logos-based tokenizer)
• Parser (recursive descent)
• Type checker (generics, traits, type inference, GATs, object safety)
• Code generator (LLVM IR via inkwell, JavaScript ESM, WASM)
• Standard library (74 modules: Vec, HashMap, String, File, Net, Async, GPU, etc.)
• Borrow checker (Non-Lexical Lifetimes, CFG-based dataflow, --strict-borrow)
• Slice types (&[T] / &mut [T] with fat pointers)
• Parallel compilation (DAG-based dependency resolution, 2-4x speedup)
• Self-hosting compiler (50,000+ LOC, 21/21 clang success, Bootstrap Phase 56)
• LSP support (diagnostics, completion, hover, go-to-definition, references, rename)
• REPL (interactive environment)
• VSCode extension + IntelliJ plugin (syntax highlighting, LSP integration)
• Optimizer (constant folding, DCE, CSE, loop unrolling, LICM, alias analysis, vectorization)
• Formatter (vaisc fmt)
• Debugger (DWARF metadata, lldb/gdb support)
• Ecosystem packages (vais-aes, vais-base64, vais-crc32, vais-csv, vais-json, vais-lz4, vais-regex, vais-sha256, vais-uuid)

Vais is designed for both compilation speed and runtime performance.

Self-Hosting Bootstrap: 50,000+ LOC, 21/21 clang compilation success (100%)

Fibonacci(35) benchmark (Apple M-series ARM64, 2026-02-11):

# Compile-time benchmarks
cargo bench -p vais-benches --bench compile_bench

# Runtime comparison benchmarks
cargo bench -p vais-benches --bench runtime_bench

The comprehensive documentation is available as an interactive mdBook site:

# Build and view the documentation
cd docs-site
./serve.sh

Visit the online documentation or browse the individual files:

• LANGUAGE_SPEC.md - Complete language specification
• STDLIB.md - Standard library reference
• TUTORIAL.md - Getting started tutorial
• Architecture.md - Compiler architecture and design
• INSTALLATION.md - Installation guide
• COVERAGE.md - Test coverage measurement guide
• MEMORY_SAFETY.md - Memory safety testing and guarantees
• ROADMAP.md - Project roadmap and progress

Vais ensures memory safety through Rust's ownership system and comprehensive testing:

# Run memory safety tests (without AddressSanitizer)
cargo test -p vaisc --test memory_safety_tests

# Run with AddressSanitizer (requires Rust nightly)
./scripts/asan-test.sh

# Run all sanitizers (ASan, UBSan, etc.)
./scripts/run-sanitizers.sh all

See MEMORY_SAFETY.md for detailed information on memory safety guarantees and testing.

brew tap vaislang/tap
brew install vais

Download from Releases (Linux, macOS Intel/ARM, Windows):

# macOS ARM
curl -LO https://github.com/vaislang/vais/releases/download/v0.1.0/vais-v0.1.0-aarch64-apple-darwin.tar.gz
tar -xzf vais-v0.1.0-aarch64-apple-darwin.tar.gz
./vais/vaisc --version

git clone https://github.com/vaislang/vais.git
cd vais && cargo build --release

docker run -it vaislang/vais:latest

After installing Vais and running your first program, here's how to continue:

# 1. Install
brew tap vaislang/tap && brew install vais

# 2. Write your first program
echo 'F main() { println("Hello, Vais!") }' > hello.vais

# 3. Run it
vaisc run hello.vais

# 4. Try the REPL
vaisc repl

Follow the structured learning path for a guided experience:

• Stage 1 (2hr): Variables, functions, control flow, structs, enums
• Stage 2 (4hr): Generics, traits, error handling, closures, stdlib
• Stage 3 (4hr): Macros, async, FFI, WASM, performance

Step-by-step project tutorials:

• CLI Tool - Build a word count tool
• HTTP Server - Create a REST API
• Data Pipeline - ETL data processing

• GitHub Discussions - Questions, ideas, show & tell
• Blog - Technical articles and language design insights
• Contributing Guide - How to contribute
• CHANGELOG - Release history
• Discord - Real-time chat (coming soon)

• GitHub: @vaislang
• Twitter/X: @vaislang
• Instagram: @vaislang

• The Self-Hosting Journey: 50,000 Lines of Vais Compiling Itself
• Vais Performance: Compilation Speed and Runtime Benchmarks
• The Design Philosophy Behind Single-Character Keywords

The prototype implementation is available on the proto branch.

MIT License