huh

On-device AI assistant for your terminal. Run huh after a failed command and it tells you what you meant to type — no internet required.

Install

pip install huh-cli
huh install   # adds shell hook to ~/.zshrc or ~/.bashrc

Then open a new terminal (or source ~/.zshrc) to activate.

Usage

# Fix your last failed command
huh

# Ask for any shell command
huh ask "compress a folder to tar.gz"
huh ask "find files modified in the last 24 hours"

# Copy suggestion to clipboard instead of running
huh --copy
huh ask "list open ports" --copy

# Show or change the model
huh model
huh model mlx-community/Qwen2.5-Coder-3B-Instruct-4bit

How it works

After huh install, your shell captures the last failed command and exit code. When you run huh, it feeds that context along with a snapshot of your environment (OS, shell, CWD, git branch) to a local LLM and returns a single corrected command.

• Apple Silicon (M1/M2/M3): uses mlx-lm with a 4-bit quantized model
• Everything else: uses llama-cpp-python with a GGUF model auto-downloaded on first use

No data leaves your machine.

Requirements

• Python 3.10+
• Apple Silicon Mac (MLX backend) or any machine that can run llama.cpp