Skip to content
/ hai Public

AI in your terminal. Run commands or ask questions - just hai it. Supports OpenAI, Claude, Gemini, DeepSeek & more.

License

Notifications You must be signed in to change notification settings

aianyai/hai

Repository files navigation

hai

Say hi to AI, right from your terminal

npm version License: GPL-3.0

hai (pronounced "hi") brings AI to your terminal. Run commands or ask questions—just hai it.

$ hai how much disk space
# → runs `df -h`, explains the result

$ hai what is Docker
# → answers your question directly

No complex setup. No switching modes. No quotes needed. Just describe what you need.

Features

  • Agent Mode - AI automatically executes shell commands based on your intent
  • Multiple AI Providers - OpenAI, Anthropic Claude, Google Gemini, and any OpenAI-compatible API
  • Profile Management - Configure multiple profiles for different models and providers
  • Interactive Mode - Multi-turn conversations in the terminal
  • Pipe Support - Process input from stdin or files
  • Predefined Prompts - Create reusable prompt templates with {{input}} placeholder
  • Thinking Mode - Enable extended thinking for supported models (Claude, o1/o3)
  • Streaming Output - Real-time response streaming with color support
  • Keyboard Shortcuts - Interrupt output with ESC or Ctrl+C

Installation

npm install -g @aiany/hai

Usage

Basic

# Single question (quotes optional)
hai What is the capital of France
hai "What is the capital of France?"

# Interactive mode
hai -i

# Interactive mode with initial message
hai -i Hello

Note: Shell special characters like ?, *, ! require quotes or escaping. Use hai "who are you?" or hai who are you\?

Agent Mode

By default, hai runs in agent mode where AI can execute shell commands to help you:

# AI will run `ls -la` to answer
hai "What files are in this directory?"

# AI will check git status and logs
hai Show me recent changes in this project

# AI will run build commands
hai Build and test this project

When a command is about to execute, you'll see a confirmation prompt:

▶ ls -la
  Execute? (Yes/no/all/cancel)
  • Y/Enter - Execute this command
  • n - Skip this command
  • a - Execute all remaining commands without asking
  • c/ESC - Cancel and stop the agent

Use -y for autonomous mode (no confirmation):

hai -y List all TypeScript files

Use --chat to disable agent mode and just chat:

hai --chat Explain what ls does

Pipe Input

# Pipe text
echo "Hello" | hai "Translate to Chinese"

# Pipe file content
cat code.js | hai "Explain this code"

File Input

# Single file
hai "Review this code" -f main.ts

# Multiple files
hai "Compare these files" -f a.ts -f b.ts

Profiles

# Use a specific profile
hai -u claude Hello

# Enable thinking mode
hai --think Explain quantum entanglement

Predefined Prompts

# Use a predefined prompt template
hai -p translate "Hello world"

Configuration

Configuration file: ~/.config/hai/settings.json

{
  "profiles": [
    {
      "name": "claude",
      "provider": "anthropic",
      "model": "claude-opus-4-5",
      "apiKey": "$ANTHROPIC_API_KEY",
      "baseURL": "https://api.openai.com/v1",
      "default": true,
      "options": {
        "think": true
      }
    },
    {
      "name": "gpt",
      "provider": "openai",
      "model": "gpt-5.2-pro",
      "apiKey": "$OPENAI_API_KEY",
      "baseURL": "https://api.openai.com/v1"
    },
    {
      "name": "gemini",
      "provider": "gemini",
      "model": "gemini-3-pro-preview",
      "apiKey": "$GOOGLE_API_KEY",
      "baseURL": "https://generativelanguage.googleapis.com/v1beta"
    },
    {
      "name": "deepseek",
      "provider": "openai-compatible",
      "model": "deepseek-reasoner",
      "apiKey": "$DEEPSEEK_API_KEY",
      "baseURL": "https://api.deepseek.com/v1"
    }
  ],
  "prompts": {
    "translate": "Translate to English: {{input}}",
    "explain": "Explain in simple terms: {{input}}"
  },
  "options": {
    "stream": true,
    "mode": "auto",
    "maxSteps": 25,
    "timeout": 120
  }
}

Options

Runtime options can be set globally or per-profile. Profile options override global options.

Option Type Default Description
stream boolean true Enable streaming output in TTY mode
think boolean false Enable thinking mode
mode string auto auto (agent) or chat (pure text)
maxSteps integer 25 Max command execution steps
timeout integer 120 Command timeout in seconds
pipe object - Pipe mode settings (stream, color)

Example with profile-specific options:

{
  "profiles": [
    {
      "name": "claude-think",
      "provider": "anthropic",
      "model": "claude-opus-4-5",
      "apiKey": "$ANTHROPIC_API_KEY",
      "options": {
        "think": true,
        "mode": "chat"
      },
      "headers": { "X-Custom-Auth": "token" },
      "providerOptions": {
        "thinking": { "type": "enabled", "budgetTokens": 20000 }
      }
    }
  ],
  "options": {
    "stream": true
  }
}

Provider Options

providerOptions accepts the same parameters as AI SDK's providerOptions. These are passed directly to the provider when calling the model.

Custom Headers

Use headers to add custom HTTP headers to API requests.

Environment Variables

Use $ENV_VAR syntax in apiKey and baseURL fields to reference environment variables:

{
  "apiKey": "$OPENAI_API_KEY",
  "baseURL": "$MY_CUSTOM_BASE_URL"
}

You can also use literal values directly:

{
  "apiKey": "sk-xxxx",
  "baseURL": "http://localhost:11434/v1"
}

Providers

Provider Description
openai OpenAI Responses API
openai-compatible OpenAI Chat Completions API
anthropic Anthropic Claude
gemini Google Gemini

CLI Options

Option Short Description
--interact -i Interactive mode
--yes -y Autonomous mode (auto-confirm)
--chat Chat mode (disable command execution)
--prompt <name> -p Use predefined prompt
--use <name> -u Use a specific profile
--file <path> -f Input file (can use multiple times)
--think Enable thinking mode
--no-think Disable thinking mode
--stream Force streaming output
--no-stream Disable streaming output
--max-steps <n> Max command execution steps (default: 25)
--timeout <secs> Command timeout in seconds (default: 120)
--debug Show full error stack traces
--config <path> Custom config file path
--version -v Show version
--help -h Show help

Keyboard Shortcuts

Key Single-shot Mode Interactive Mode Command Confirm
Ctrl+C Exit Exit Exit
ESC Stop output Stop current output Cancel
Y/Enter - - Execute
n - - Skip
a - - Execute all
c - - Cancel

In interactive mode, pressing ESC interrupts the current response but allows you to continue the conversation. Use Ctrl+C or type /exit to quit.

中文文档

查看中文 README

License

GPL-3.0