Sabi-TUI

A terminal-based AI agent implementing the ReAct (Reasoning + Acting) pattern for system administration. Describe tasks in natural language, review AI-generated shell commands, and get analysis of results.

Features

• 🤖 Multi-provider AI - Gemini, OpenAI, Ollama, Groq, Together AI
• 💻 Terminal access - Execute commands with safety checks
• 🐍 Python executor - Run Python code for calculations (auto-detected)
• 🖼️ Image analysis - Paste images from clipboard or file for AI analysis
• 🔌 MCP Support - Extend with Model Context Protocol servers
• 🔒 Safe mode - Preview commands without execution
• 💾 Multi-session - Save and switch between conversation sessions
• 🛡️ 2-step confirmation - Dangerous commands require explicit confirmation
• 🚫 Interactive command blocking - Prevents hanging on vim, ssh, etc.

Installation

Quick Install (Recommended)

curl -sSL https://raw.githubusercontent.com/n4ar/sabi-tui/main/setup.sh | bash

Manual Download

Download from Releases:

Build from Source

git clone https://github.com/n4ar/sabi-tui.git
cd sabi-tui
cargo build --release
cp target/release/sabi ~/.local/bin/

Quick Start

Run sabi and follow the onboarding wizard:

🚀 Welcome to Sabi-TUI!

Select provider:
  1) Gemini (Google AI)
  2) OpenAI
  3) OpenAI-compatible (Ollama, Groq, Together, etc.)

Choice [1]: 

Configuration

All config stored in ~/.sabi/:

# ~/.sabi/config.toml

# Provider: "gemini" or "openai"
provider = "gemini"
api_key = "your-api-key"
model = "gemini-2.5-flash"

# For OpenAI-compatible APIs (Ollama, Groq, etc.)
# provider = "openai"
# base_url = "http://localhost:11434/v1"
# model = "llama3.2"

Provider Examples

# Gemini
provider = "gemini"
api_key = "your-gemini-key"
model = "gemini-2.5-flash"

# OpenAI
provider = "openai"
api_key = "sk-xxx"
model = "gpt-4o"

# Ollama (local)
provider = "openai"
base_url = "http://localhost:11434/v1"
model = "llama3.2"

# Groq
provider = "openai"
base_url = "https://api.groq.com/openai/v1"
api_key = "gsk_xxx"
model = "llama-3.3-70b-versatile"

Usage

sabi                    # Interactive TUI mode
sabi -q "prompt"        # Quick query (text response)
sabi -x "prompt"        # Execute mode (with confirmation)
sabi --safe             # Safe mode (preview only)
sabi mcp <command>      # Manage MCP servers
sabi --version          # Show version
sabi --help             # Show help

Quick CLI Mode

Get instant answers without entering the TUI:

# Quick query - get text response
sabi -q "what is my IP address"

# Execute mode - run command with confirmation dialog
sabi -x "list large files in current directory"

Execute mode (-x) shows:

1. Confirmation dialog - Review command before execution
2. Result dialog - Output with AI summary

Slash Commands

Press Tab to autocomplete commands.

Shell Escape

Use ! prefix to run shell commands directly without AI:

!ls -la          # List files
!pwd             # Print working directory
!cat file.txt    # View file contents

Keybindings

MCP (Model Context Protocol)

Extend Sabi with external tools via MCP servers (stdio or HTTP).

Managing MCP Servers

# Add stdio server
sabi mcp add <name> <command> [args...]

# Add HTTP server
sabi mcp add -t http -H "API-KEY: xxx" <name> <url>

# Set environment variable (stdio)
sabi mcp env <name> KEY=VALUE

# List configured servers
sabi mcp list

# Remove a server
sabi mcp remove <name>

Examples

# Filesystem server (stdio)
sabi mcp add filesystem npx -y @modelcontextprotocol/server-filesystem /home

# Git server (stdio)
sabi mcp add git npx -y @modelcontextprotocol/server-git

# Brave Search with API key (stdio)
sabi mcp add brave npx -y @anthropic/mcp-server-brave-search
sabi mcp env brave BRAVE_API_KEY=your-api-key

# Context7 (HTTP)
sabi mcp add -t http -H "CONTEXT7_API_KEY: your-key" context7 https://mcp.context7.com/mcp

# HTTP server with multiple headers
sabi mcp add -t http -H "Authorization: Bearer xxx" -H "X-Custom: value" myserver https://example.com/mcp

MCP Configuration File

Servers are stored in ~/.sabi/mcp.toml. See examples/mcp.toml.example for a complete example.

Safety Features

🛡️ 2-Step Confirmation for Dangerous Commands

Commands targeting sensitive paths (~, /Users, /etc) or using destructive patterns (rm -rf) require:

1. First Enter - Warning displayed
2. Second Enter - Must type "I understand the risks"

⚠️ DANGEROUS COMMAND DETECTED!
This command could cause irreversible damage.
Press Enter again to proceed to final confirmation.

🛑 FINAL CONFIRMATION REQUIRED
Type exactly: I understand the risks

⛔ Unknown Tool Blocking

AI cannot create arbitrary tools. Only allowed:

• run_cmd - Shell commands
• run_python - Python code
• read_file / write_file - File operations
• search - File search
• mcp - MCP server tools

🚫 Dangerous Path Detection

Operations on these paths trigger safety checks:

• Home directories: ~, /Users, /home, /root
• System directories: /etc, /var, /usr, /bin, /sbin
• macOS system: /System, /Library, /Applications

Available Tools

Troubleshooting

"API key not found"

Run sabi to start onboarding, or edit ~/.sabi/config.toml

Python not detected

Install Python 3: brew install python3 (macOS) or apt install python3 (Linux)

Model not working

Use /model to list available models and switch

MCP server not starting

• Check if the command exists: npx -y @modelcontextprotocol/server-filesystem --help
• Verify environment variables: sabi mcp list
• Check server logs in stderr

Uninstall

# Using uninstall script
curl -sSL https://raw.githubusercontent.com/n4ar/sabi-tui/main/uninstall.sh | bash

# Or manually
rm ~/.local/bin/sabi
rm -rf ~/.sabi

License

MIT