grok-cli: an open-source coding agent for the Grok API

Disclaimer: This project is community-built, open-source, and not affiliated with, endorsed by, or sponsored by xAI Corp. "Grok" is a trademark of xAI Corp. This tool uses the publicly available Grok API.

An open-source terminal coding agent that connects to xAI’s Grok API — real-time X search, web search, the full Grok model lineup, sub-agents on by default, remote control via Telegram (pair once, drive the agent from your phone while the CLI runs), and a terminal UI built with Bun and OpenTUI.

https://github.com/user-attachments/assets/7ca4f6df-50ca-4e9c-91b2-d4abad5c66cb

Install

curl -fsSL https://raw.githubusercontent.com/superagent-ai/grok-cli/main/install.sh | bash

Alternative installs (requires Bun on PATH):

bun add -g grok-dev

Self-management (script-installed only):

grok update
grok uninstall
grok uninstall --dry-run
grok uninstall --keep-config

Prerequisites: a Grok API key from x.ai and a modern terminal emulator for the interactive OpenTUI experience. Headless --prompt mode does not depend on terminal UI support. If you want host desktop automation via the built-in computer sub-agent, also enable Accessibility permission for your terminal app on macOS.

Run it

Interactive (default) — launches the OpenTUI coding agent:

grok

Supported terminals

For the most reliable interactive OpenTUI experience, use a modern terminal emulator. We currently document and recommend:

• WezTerm (cross-platform)
• Alacritty (cross-platform)
• Ghostty (macOS and Linux)
• Kitty (macOS and Linux)

Other modern terminals may work, but these are the terminal apps we currently recommend and document for interactive use.

Pick a project directory:

grok -d /path/to/your/repo

Headless — one prompt, then exit (scripts, CI, automation):

grok --prompt "run the test suite and summarize failures"
grok -p "show me package.json" --directory /path/to/project
grok --prompt "refactor X" --max-tool-rounds 30
grok --prompt "summarize the repo state" --format json
grok --prompt "review the repo overnight" --batch-api
grok --verify

--batch-api uses xAI's Batch API for lower-cost unattended runs. It is a good fit for scripts, CI, schedules, and other non-interactive workflows where a delayed result is fine.

Continue a saved session:

grok --session latest
grok -s <session-id>

Works in interactive mode too—same flag.

Structured headless output:

grok --prompt "summarize the repo state" --format json

--format json emits a newline-delimited JSON event stream instead of the default human-readable text output. Events are semantic, step-level records such as step_start, text, tool_use, step_finish, and error.

Computer sub-agent

Grok ships a built-in **computer** sub-agent backed by [agent-desktop](https://github.com/lahfir/agent-desktop) for host desktop automation on macOS.

Ask for it in natural language, for example:

grok "Use the computer sub-agent to take a screenshot of my host desktop and tell me what is open."
grok "Use the computer sub-agent to launch Google Chrome, snapshot the UI, and tell me which refs correspond to the address bar and tabs."

Notes:

• Screenshots are saved under **.grok/computer/** by default.
• The primary workflow is snapshot -> refs -> action -> snapshot using agent-desktop accessibility snapshots and stable refs like @e1.
• computer_screenshot is available for visual confirmation, but the preferred path is computer_snapshot plus ref-based actions such as computer_click, computer_type, and computer_scroll.
• macOS requires System Settings → Privacy & Security → Accessibility access for the terminal app running grok.
• agent-desktop currently targets macOS.
• If Bun blocks the native binary download during install, run:

node ./node_modules/agent-desktop/scripts/postinstall.js

Scheduling

Schedules let Grok run a headless prompt on a recurring schedule or once. Ask for it in natural language, for example:

Create a schedule named daily-changelog-update that runs every weekday at 9am
and updates CHANGELOG.md from the latest merged commits.

Recurring schedules require the background daemon:

grok daemon --background

Use /schedule in the TUI to browse saved schedules. One-time schedules start immediately in the background; recurring schedules keep running as long as the daemon is active.

List Grok models and pricing hints:

grok models

Pass an opening message without another prompt:

grok fix the flaky test in src/foo.test.ts

Generate images or short videos from chat:

grok "Generate a retro-futuristic logo for my CLI called Grok Forge"
grok "Edit ./assets/hero.png into a watercolor poster"
grok "Animate ./assets/cover.jpg into a 6 second cinematic push-in"

Image and video generation are exposed as agent tools inside normal chat sessions. You keep using a text model for the session, and Grok saves generated media under .grok/generated-media/ by default unless you ask for a specific output path.

What you actually get

Coming soon

Deeper autonomous agent testing — persistent sandbox sessions, richer browser workflows, and stronger "prove it works" evidence.

API key (pick one)

Environment (good for CI):

export GROK_API_KEY=your_key_here

**.env** in the project (see .env.example if present):

GROK_API_KEY=your_key_here

CLI once:

grok -k your_key_here

Saved in user settings — ~/.grok/user-settings.json:

{ "apiKey": "your_key_here" }

Optional **subAgents** — custom foreground sub-agents. Each entry needs **name**, **model**, and **instruction**:

{
  "subAgents": [
    {
      "name": "security-review",
      "model": "grok-4.3",
      "instruction": "Prioritize security implications and suggest concrete fixes."
    }
  ]
}

Names cannot be general, explore, vision, verify, or computer because those are reserved for the built-in sub-agents.

Optional: **GROK_BASE_URL** (default https://api.x.ai/v1), **GROK_MODEL**, **GROK_MAX_TOKENS**.

Telegram (remote control) — short version

1. Create a bot with @BotFather, copy the token.
2. Set **TELEGRAM_BOT_TOKEN** or add **telegram.botToken** in ~/.grok/user-settings.json (the TUI **/remote-control** flow can save it).
3. Start **grok**, open **/remote-control** → Telegram if needed, then in Telegram DM your bot: **/pair, enter the 6-character code** in the terminal when asked.
4. First user must be approved once; after that, it’s remembered. Keep the CLI process running while you use the bot (long polling lives in that process).

Voice & audio messages

Send a voice note or audio attachment in Telegram and Grok will transcribe it with the Grok Speech-to-Text API (POST https://api.x.ai/v1/stt) before passing the text to the agent. The endpoint accepts Telegram's OGG/Opus voice notes and common audio containers (MP3, WAV, M4A, FLAC, AAC) directly — no local model download, whisper-cli, or ffmpeg required.

Prerequisites

• A valid GROK_API_KEY (the same key used for the agent). Transcription reuses the CLI's apiKey / baseURL resolution, so if the agent can reach xAI, transcription will too.

Configure in ~/.grok/user-settings.json

{
  "telegram": {
    "botToken": "YOUR_BOT_TOKEN",
    "audioInput": {
      "enabled": true,
      "language": "en"
    }
  }
}

Optional headless flow when you do not want the TUI open:

grok telegram-bridge

Treat the bot token like a password.

Hooks

Hooks execute shell commands at key agent lifecycle events — enforce policies, run linters, trigger tests, or log activity.

Configure in ~/.grok/user-settings.json:

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "bash",
        "hooks": [
          {
            "type": "command",
            "command": "./scripts/lint-before-edit.sh",
            "timeout": 10
          }
        ]
      }
    ]
  }
}

Hook commands receive JSON on stdin (event details) and can return JSON on stdout. Exit code 0 = success, 2 = block the action, other = non-blocking error.

Supported events: PreToolUse, PostToolUse, PostToolUseFailure, UserPromptSubmit, SessionStart, SessionEnd, Stop, StopFailure, SubagentStart, SubagentStop, TaskCreated, TaskCompleted, PreCompact, PostCompact, Notification, InstructionsLoaded, CwdChanged.

Instructions & project brain

• **AGENTS.md — merged from git root down to your cwd (Codex-style; see repo docs). `AGENTS.override.md**` wins per directory when present.

Project settings

Project file: **.grok/settings.json** — e.g. the current model for this project.

Sandbox

Grok CLI can run shell commands inside a Shuru microVM sandbox so the agent can't touch your host filesystem or network.

Requires macOS 14+ on Apple Silicon.

Enable it with --sandbox on the CLI, or toggle it from the TUI with /sandbox.

On the first interactive run in a new directory, Grok asks whether to remember sandbox or host mode for that workspace and stores the choi