AI-Shell-Team/aish

What it solves

AISH is an AI-native interactive shell designed for system operations and troubleshooting. It bridges the gap between natural language requests and terminal execution, allowing users to perform complex system tasks without needing to memorize every specific command flag or syntax.

How it works

The project implements a full Pseudo-Terminal (PTY) to support interactive programs like vim or ssh while integrating an AI handler that triggers when a user starts a command with a semicolon (;). It uses a ReAct diagnostic agent to generate and explain commands. To prevent accidental system damage, it features a security policy engine that grades AI-generated commands by risk level (Low, Medium, High) and can optionally run commands in a sandbox for pre-execution assessment.

Who it’s for

System administrators, DevOps engineers, and Linux users who want to automate operations and troubleshoot systems using natural language while maintaining strict control over command execution.

Highlights

• Full PTY Support: Unlike simple CLI wrappers, it runs fully interactive terminal applications.
• Risk-Based Security: Implements a confirmation flow based on risk grading and optional sandbox pre-runs.
• Multi-Model Flexibility: Fully open support for various LLMs via OpenAI-compatible APIs.
• Extensible Skills: A plugin system that allows hot-loading of new domain knowledge and workflows.
• Crate-based Architecture: Built in Rust for performance and reliability across 14 specialized crates.