aish - AI Shell Helper

A fast, terminal-based tool that converts natural language descriptions into executable shell commands using local LLMs via Ollama.

Purpose

aish eliminates the need to remember complex shell syntax by letting you describe what you want to do in plain English. It generates the command, lets you review it, and gives you the option to execute, copy, or refine it.

Key Features:

• ⚡ Instant startup - Built in Go for zero-latency execution
• 🔒 Fully local - Runs entirely on your machine via Ollama
• 🎯 macOS optimized - Designed for Zsh on macOS
• 🔄 Iterative refinement - Chat with the AI to adjust commands
• 📋 Clipboard integration - Copy commands with one keystroke
• 🚀 Single binary - No dependencies, just run it

Prerequisites

• macOS running Zsh
• Ollama installed and running
• The llama3.2:3b model (or customize in config)

# Install Ollama (if not already installed)
curl -fsSL https://ollama.ai/install.sh | sh

# Pull the default model
ollama pull llama3.2:3b

Installation

git clone https://github.com/davide-parini/aish
cd aish
./install.sh

The install script will:

• Build the binary
• Install to ~/.local/bin (no sudo required)
• Delete any existing config for a fresh start
• Remind you to add ~/.local/bin to PATH if needed

Usage

aish <your goal in natural language>

Interactive Mode

After generating a command, simply press a key to choose your action:

• [⏎] - Execute the command immediately 🚀
• [1] - Copy to clipboard and exit 📋
• [2] - Refine the command with additional instructions ✨
• [3] - Explain the command in detail 💡
• [q] - Exit without action 👋

Example Session:

🧠 Thinking...find . -type f -size +100M
[⏎] Execute 🚀
[1] Copy 📋
[2] Refine ✨
[3] Explain 💡
[q] Exit 👋

Refinement:

Press [2] to refine the command:

✨ Refinement prompt: only show files modified in the last week

The AI combines your refinement with the previous command, maintaining context across iterations.

Explanation:

Press [3] for a detailed breakdown:

💡 Explanation:
This command searches for files larger than 100 megabytes.
The find utility starts from the current directory (.)...

Explanations are generated in a separate session and won't affect refinement context.

Configuration

On first run, aish creates ~/.config/aish/config.json with default settings:

{
  "ollama_url": "http://localhost:11434",
  "model": "llama3.2:3b",
  "system_prompt": "..."
}

Customization:

• Change ollama_url if running Ollama remotely
• Use a different model (e.g., mistral, codellama)
• Modify system_prompt for different behavior

How It Works

1. Input: You describe what you want in natural language
2. Generation: aish sends your request to Ollama with a specialized prompt
3. Review: The generated command is displayed for your approval
4. Action: Execute, copy, or refine based on your needs

The system prompt is engineered to output raw, executable commands without markdown formatting or explanations, ensuring compatibility with direct execution.

Requirements

• Go: 1.19+ (for building from source)
• RAM: ~8GB for llama3.2:3b model
• Disk: ~2GB for model storage

License

MIT

Contributing

Contributions welcome! Feel free to open issues or submit pull requests.