A terminal UI for xAI's Grok models. Interactive chat, streaming responses, live web search, slash commands, and full CLI options — all from your terminal.

Built with Python, talking directly to the xAI API (no OpenAI compatibility wrapper).

• Streaming responses with live-updating Markdown rendering
• xAI Live Search — toggle real-time web data in responses (/search)
• Reasoning display — see Grok's thinking process on mini models (/reasoning)
• Slash commands — 16 built-in commands for controlling the session
• One-shot mode — pipe prompts or pass them inline
• Conversation management — history, export, compact (AI-summarized)
• Rich terminal output — syntax highlighting, markdown, panels
• Tab completion — slash commands auto-complete
• Input history — arrow keys recall previous prompts (persisted across sessions)
• Configurable — model, temperature, max tokens, system prompt, all adjustable at runtime

• Python 3.10+
• xAI API key — get one at console.x.ai

Note: This uses the xAI API, which is separate from a SuperGrok subscription. API usage is billed per-token at x.ai/api/pricing.

git clone https://github.com/Norlem/grok-cli.git
cd grok-cli
python3 -m venv .venv
source .venv/bin/activate
pip install -e .

pip install git+https://github.com/Norlem/grok-cli.git

grok --init

This walks you through:

1. Entering your xAI API key
2. Choosing a default model
3. Enabling/disabling live web search

Settings are saved to ~/.grok-cli/config.json.

Alternatively, set your API key as an environment variable:

export XAI_API_KEY="xai-your-key-here"

Add this to your ~/.zshrc or ~/.bashrc to persist it.

grok

This opens the interactive TUI. Type messages, use slash commands, and chat with Grok.

Pass a prompt directly — Grok responds and exits:

# Inline argument
grok "what is the mass of the sun?"

# With the -p flag
grok -p "explain quantum entanglement in simple terms"

# With options
grok -m grok-3-mini --search "latest SpaceX launch news"

# Pipe from stdin
echo "summarize this error log" | grok
cat error.log | grok -p "what went wrong?"

All commands work in interactive mode. Type /help to see them at any time.

Switch models at any time:

/model grok-3-mini

Or list what's available from xAI:

/model

Grok can search the web in real-time — unique to xAI's API. Toggle it:

/search

Or enable from the command line:

grok --search "who won the game last night?"

When enabled, Grok will pull live data from the web to inform its responses.

For mini models, you can control how much "thinking" Grok does before responding:

/reasoning high

Options: low, medium, high, off

When reasoning is active during streaming, you'll see a "thinking" indicator before the response appears.

Long conversations eat context. The /compact command asks a fast model to summarize the entire conversation, replacing it with a 2-message summary:

/compact

This keeps the essential context while freeing up token budget.

Settings are stored in ~/.grok-cli/config.json:

{
  "api_key": "xai-...",
  "model": "grok-3",
  "temperature": 0.7,
  "max_tokens": 4096,
  "system_prompt": "You are Grok, a helpful AI assistant built by xAI.",
  "stream": true,
  "live_search": false,
  "show_tokens": false,
  "theme": "monokai",
  "save_conversations": true,
  "max_history_messages": 100,
  "reasoning_effort": ""
}

Edit directly or use /save-config to persist runtime changes.

Prompt history (arrow key recall) is saved to ~/.grok-cli/prompt_history.

# Quick question
grok "how do I reverse a linked list in Rust?"

# Use the mini model for fast answers
grok -m grok-3-mini-fast "what's 2+2?"

# Research with live search
grok --search "summarize today's tech news"

# Code review from a file
cat main.py | grok -s "You are a senior code reviewer." -p "review this code"

# Low temperature for deterministic output
grok -t 0.1 -p "translate 'hello world' to Japanese, Korean, and Mandarin"

# Interactive with custom system prompt
grok -s "You are a Socratic tutor. Answer questions with questions."

MIT