English | 简体中文

---

AISH

Empower the Shell to think. Evolve Operations.

A Real AI Shell: Complete PTY + Configurable Security & Risk Control

---

Table of Contents

• Why Choose AISH
• Quick Start
• Installation
• Uninstallation
• Configuration
• Usage
• Security & Risk Control
• Skills (Plugins)
• Data & Privacy
• Documentation
• Community & Support
• Development & Testing
• Contributing
• License

---

Why Choose AISH

• True Interactive Shell: Full PTY support, runs interactive programs like vim / ssh / top
• AI Native Integration: Describe tasks in natural language, generate, explain and execute commands
• Secure & Controllable: AI commands have risk grading and confirmation flow; optional sandbox pre-run for change assessment
• Extensible: Skills plugin system with hot loading and priority override
• Low Migration Cost: Compatible with regular commands and workflows, everything in terminal by default

---

Feature Comparison

Feature	AISH	Claude Code
🎯 Core Positioning	Ops/System Troubleshooting CLI	Development Coding Assistant
🤖 Multi-Model Support	✅ Fully Open	⚠️ Mainly Claude
🔧 Sub-Agent System	✅ ReAct Diagnostic Agent	✅ Multiple Agent Types
🧩 Skills Support	✅ Hot Loading	✅
🖥️ Native Terminal Integration	✅ Full PTY Support	⚠️ Limited Support
🛡️ Security Risk Assessment	✅ Security Confirmation	✅ Security Confirmation
🌐 Local Model Support	✅ Fully Supported	Fully Supported
📁 File Operation Tools	✅ Minimal Essential Suppport	✅ Full Support
💰 Completely Free	✅ Open Source	❌ Paid Service
📊 Observability	✅ Langfuse Optional	⚠️ Built-in
🌍 Multi-language Output	✅ Auto Detection	✅

---

Quick Start

1) Install and Launch

Option 1: One-line install (Recommended)

curl -fsSL https://www.aishell.ai/repo/install.sh | bash

Option 2: Manual bundle install

Download the matching aish-<version>-linux-<arch>.tar.gz bundle from the official release directory, then run:

tar -xzf aish-<version>-linux-<arch>.tar.gz
cd aish-<version>-linux-<arch>
sudo ./install.sh

Then launch:

aish

Note: aish without subcommands is equivalent to aish run.

2) Use Like a Regular Shell

aish> ls -la
aish> cd /etc
aish> vim hosts

3) Let AI Do the Work (Start with ;)

Starting with ; or ； enters AI mode:

aish> ;find files larger than 100M in current directory and sort by size
aish> ;explain this command: tar -czf a.tgz ./dir

---

Installation

Linux Release Bundle

curl -fsSL https://www.aishell.ai/repo/install.sh | bash

The installer resolves the latest stable version, downloads the matching bundle for your architecture, and installs aish, aish-sandbox, and aish-uninstall into /usr/local/bin.

Run from Source (Development/Trial)

cargo run --bin aish

---

Uninstallation

Uninstall (keep configuration files):

sudo aish-uninstall

Complete uninstallation (also removes system-level security policies):

sudo aish-uninstall --purge-config

Optional: Clean user-level configuration (will clear model/API keys etc.):

rm -rf ~/.config/aish

---

Configuration

Configuration File Location

• Default: ~/.config/aish/config.yaml (or $XDG_CONFIG_HOME/aish/config.yaml if XDG_CONFIG_HOME is set)

Priority (High to Low)

1. Command-line arguments
2. Environment variables
3. Configuration file

Minimal Configuration Example

# ~/.config/aish/config.yaml
model: openai/deepseek-chat
api_base: https://openrouter.ai/api/v1
api_key: your_api_key

Alternatively via environment variables (more suitable for secrets):

export AISH_MODEL="openai/deepseek-chat"
export AISH_API_BASE="https://openrouter.ai/api/v1"
export AISH_API_KEY="your_api_key"

Tip: LiteLLM also supports reading vendor-specific environment variables (e.g., OPENAI_API_KEY, ANTHROPIC_API_KEY).

Interactive configuration (optional):

aish setup

Tool calling compatibility check (confirm selected model/channel supports tool calling):

aish check-tool-support --model openai/deepseek-chat --api-base https://openrouter.ai/api/v1 --api-key your_api_key

Langfuse (optional observability):

1. Enable in configuration:

enable_langfuse: true

2. Set environment variables:

export LANGFUSE_PUBLIC_KEY="..."
export LANGFUSE_SECRET_KEY="..."
export LANGFUSE_HOST="https://cloud.langfuse.com"

aish check-langfuse will run checks when check_langfuse.py exists in project root.

---

Usage

Common Input Types

Type	Example	Description
Shell Commands	ls -la, cd /path, git status	Execute regular commands directly
AI Requests	;how to check port usage, ;find files larger than 100M	Enter AI mode with ;/； prefix
Built-in Commands	help, clear, exit, quit	Shell built-in control commands
Model Switching	/model gpt-4	View or switch model

Shell Compatibility (PTY)

aish> ssh user@host
aish> top
aish> vim /etc/hosts

---

Security & Risk Control

AI Shell only performs security assessment on AI-generated and ready-to-execute commands.

Risk Levels

• LOW: Allowed by default
• MEDIUM: Confirmation before execution
• HIGH: Blocked by default

Security Policy File Path

Policy files are resolved in this order:

1. /etc/aish/security_policy.yaml (system-level)
2. ~/.config/aish/security_policy.yaml (user-level; auto-generated template if not exists)

Sandbox Pre-run (Optional, Recommended for Production)

Default policy has sandbox pre-run disabled. To enable:

1. Set in security policy:

global:
  enable_sandbox: true

2. Start privileged sandbox service (systemd):

sudo systemctl enable --now aish-sandbox.socket

Default socket: /run/aish/sandbox.sock. When sandbox is unavailable, it will fallback according to sandbox_off_action (BLOCK/CONFIRM/ALLOW) in policy.

---

Skills (Plugins)

Skills extend AI's domain knowledge and workflows, supporting hot loading and priority override.

Default scan directories and priority:

• ~/.config/aish/skills/ (or $AISH_CONFIG_DIR/skills)
• ~/.claude/skills/

Packaged versions will attempt to copy system-level skills to user directory on first launch (e.g., /usr/share/aish/skills).

For more details, see: docs/skills-guide.md

---

Data & Privacy

This project stores the following data locally (for troubleshooting and traceability):

• Logs: Default ~/.config/aish/logs/aish.log
• Sessions/History: Default ~/.local/share/aish/sessions.db (SQLite)
• Large Output Offload: Default ~/.local/share/aish/offload/

Recommendations:

• Don't commit real API keys to repository; prefer environment variables or secret management systems.
• Production environments can combine security policies to limit AI-accessible directory scope.

---

Documentation

• Configuration Guide: CONFIGURATION.md
• Quick Start: QUICKSTART.md
• Skills Usage: docs/skills-guide.md
• Command Correction Mechanism: docs/command-interaction-correction.md

---

Community & Support

Link	Description
Official Website	Project homepage and more information
GitHub Repository	Source code and issue tracking
GitHub Issues	Bug reports
GitHub Discussions	Community discussions
Discord	Join the community
LINUX DO	Community friend link

If you find AISH helpful, please consider giving us a Star ⭐ It helps more people discover the project and keeps us motivated to continue improving it.

---

Development & Testing

This repository is now the Rust implementation of aish. The remaining Python scripts under packaging/scripts/ are release automation helpers and are covered by shell smoke tests.

# Build
cargo build --release

# Run Rust tests
cargo test --workspace

# Lint Rust code
cargo clippy --all-targets -- -D warnings

# Run packaging release script smoke tests
./packaging/tests/release_scripts_smoke.sh

Architecture (14 crates)

Crate	Description
aish-core	Error types, shared enums (RiskLevel, MemoryCategory, etc.)
aish-config	YAML config loading with XDG paths and env overrides
aish-i18n	Internationalization with 6 locales and embedded fallback
aish-pty	PTY executor with fork/exec, select()-based I/O, output offload
aish-llm	OpenAI-compatible HTTP client with SSE streaming and tool calling
aish-session	SQLite session persistence with WAL mode
aish-context	Sliding-window context manager with tiktoken token counting
aish-security	Security policy engine with glob-to-regex pattern matching
aish-skills	Skill plugin discovery with hot-reload via notify
aish-memory	Markdown-based long-term memory with relevance scoring
aish-tools	Built-in tools: bash, fs (read/write/edit), ask_user, memory, skill
aish-scripts	.aish script system with frontmatter, ai "prompt" syntax, hooks
aish-shell	Main shell: REPL loop, AI handler, built-in commands, animation
aish-cli	CLI entry point with clap derive macros

---

Contributing

See CONTRIBUTING.md for guidelines.

License

LICENSE (Apache 2.0)