Talk to your terminal.

Transform natural language into safe, POSIX-compliant shell commands. Powered by local LLMs. Privacy-first. Blazingly fast on Apple Silicon.

Get Started View on GitHub

Safety First

Comprehensive validation blocks dangerous commands before execution. Pattern matching for rm -rf, fork bombs, and more.

Blazingly Fast

Optimized for Apple Silicon with MLX. Sub-2-second inference on M1/M2/M3/M4 chips with GPU acceleration.

Privacy First

All inference runs locally on your machine. Your commands and data never leave your computer.

POSIX Compliant

Generated commands work across all Unix systems. Standard utilities like find, grep, awk, and sed.

Quick Start

# Install via cargo
cargo install caro

# Or with Homebrew (macOS)
brew install wildcard/tap/caro

# Convert natural language to shell commands
caro "find all rust files modified in the last week"
# → find . -name "*.rs" -mtime -7

caro "show disk usage sorted by size"
# → du -sh * | sort -hr

caro "count lines in all python files"
# → find . -name "*.py" -exec wc -l {} + | tail -1

# Set your preferred backend
export CARO_BACKEND=mlx    # Apple Silicon
export CARO_BACKEND=ollama # Any platform

# Optional: customize model
export CARO_MODEL=codellama

Why caro?

No Cloud Dependency

Everything runs locally. No API keys, no rate limits, no privacy concerns. Works offline after the initial model download.

Single Binary

One Rust binary under 50MB. No Python environments, no containers, no complex dependencies to manage.

Multiple Backends

Supports MLX for Apple Silicon, Ollama for local inference, and vLLM for high-performance serving.

Developer Friendly

Intuitive CLI with colored output, confirmation prompts, and helpful error messages.

Next Steps

Installation Guide Install caro and get running in minutes

Quick Start Tutorial Learn the basics with practical examples

macOS Setup Optimize for Apple Silicon with MLX GPU acceleration