Ghostclaw

an AI that lives on your computer and does stuff for you. Public beta.

Last verified:

Visit Ghostclaw

What is Ghostclaw?

GhostClaw is an open-source personal AI agent that turns a spare computer into an autonomous AI co-worker with full system access. It is a bare-metal fork of NanoClaw with containers stripped out, providing ~4K lines of code compared to OpenClaw's 500K lines. The tool lets AI agents run wild on your machine with unrestricted access to Bash, files, web search, and email without sandboxing or permission prompts.

The platform is Telegram-first, meaning you interact with your AI agent through Telegram DMs just like a co-worker. You can send voice notes, photos, and use group chats without installing any additional app. It includes a built-in Mission Control dashboard at localhost:3333 for real-time logs, task management, chats, and research monitoring. The agent supports scheduled tasks like morning briefings, overnight research, and cron jobs that work while you sleep.

GhostClaw features an infinitely customizable skills system where you add capabilities by typing commands like /add-gmail-agent or /add-voice-transcription. Each skill is a markdown file requiring no code changes, config files, or rebuilds. Built-in skills include Gmail agent, voice transcription with ElevenLabs, voice reply, morning briefing, heartbeat health checks, Telegram swarm for multi-agent teams, PR babysitter for CI monitoring, and autonomous overnight task loops.

The tool is designed for developers, tech enthusiasts, and anyone with a spare computer who wants a personal AI agent with genuine machine access. It requires Node.js 20+, an Anthropic API key (Claude Max or API keys), and runs on macOS or Linux. Setup takes approximately 10 minutes with a one-command installer that guides you through API key configuration, Telegram bot setup, personality selection, and auto-start configuration.

Ghostclaw pricing

Pricing model: Freemium

GhostClaw itself is completely free and open source under the MIT licence. The only cost is requiring an Anthropic API key for Claude (Claude Max or API keys), which is a paid service from Anthropic. Some features like voice transcription and voice reply use ElevenLabs Scribe and TTS which likely require separate ElevenLabs subscriptions. The one-command installer is free and the software runs on your own hardware with no licensing fees.

Ghostclaw pros

  • Open source with MIT licence - fully forkable and customizable
  • Only ~4K lines of code - minimal and maintainable compared to OpenClaw's 500K
  • Bare-metal execution with full machine access, no Docker containers
  • No sandboxing or permission prompts blocking agent actions
  • Telegram-first interface - no app installation required
  • Supports voice notes, photos, and group chats naturally
  • 449 tests passing ensuring code reliability
  • Built-in Mission Control dashboard at localhost:3333
  • Real-time logs, task tracking, and chat monitoring in dashboard
  • Skills system adds features via markdown files without code changes
  • Switch AI models (Sonnet, Opus, Haiku) instantly from Telegram with /model
  • Built-in Gmail agent with OAuth setup included
  • Voice transcription and voice reply with ElevenLabs integration
  • Scheduled tasks work overnight while you sleep
  • Telegram swarm enables multi-agent teams with separate bot identities
  • PR babysitter automatically monitors failing CI and unanswered reviews
  • Autonomous overnight task loop (run-ralph) commits after each task
  • One-command installer walks through entire setup process
  • 10 minutes to complete setup from start to finish
  • Heartbeat skill provides periodic health checks with minimal alerts

Ghostclaw cons

  • Requires paid Anthropic API key (Claude Max or API keys)
  • Only supports macOS and Linux - no Windows support
  • Requires Node.js 20+ installed first
  • Needs a spare dedicated computer to run continuously
  • No sandboxing means full system access could be risky if compromised
  • Telegram dependency - must use Telegram for all interactions
  • ElevenLabs integrations likely require separate paid subscriptions
  • Relatively new project with smaller community than established alternatives

Frequently asked questions about Ghostclaw

What is GhostClaw?

GhostClaw is an open-source personal AI agent that is a bare-metal fork of NanoClaw with containers stripped out. It gives your spare computer its own accounts and lets an AI agent run wild with full system access. It's approximately 4K lines of code, Telegram-first, and takes 10 minutes to set up.

How is GhostClaw different from OpenClaw and NanoClaw?

OpenClaw has 500K lines of code with containers and a config maze that's overkill for personal use. NanoClaw is clean and minimal but cages agents in Docker containers. GhostClaw takes NanoClaw's core, rips out the containers, and provides full machine access at ~4K lines - combining NanoClaw's simplicity with OpenClaw's freedom without the mess.

How do I install GhostClaw?

Run the one-command installer: curl -fsSL https://ghostclaw.io/install.sh | bash. The installer walks you through everything including API key setup, Telegram bot configuration, personality selection, and auto-start setup. You need Node.js 20+ and an Anthropic API key beforehand.

What operating systems does GhostClaw support?

GhostClaw runs on macOS or Linux only. It does not support Windows. You also need Node.js 20+ installed on your system.

How do I interact with my GhostClaw agent?

You interact with GhostClaw through Telegram DMs like a co-worker. Send voice notes, photos, join group chats, and type commands. No separate app installation is required - everything works through Telegram's native interface.

What skills can I add to GhostClaw?

You can add Gmail agent (/add-gmail-agent), voice transcription (/add-voice-transcription), voice reply (/add-voice-reply), morning briefing (/add-morning-briefing), heartbeat health checks (/add-heartbeat), Telegram swarm for multi-agent (/add-telegram-swarm), PR babysitter (/pr-babysitter), and autonomous overnight loops (/run-ralph). Skills are markdown files - you can write your own or ask Claude to build one.

How does the skills system work?

Each skill is a markdown file that adds capabilities without code changes, config files, or rebuilds. Just type a command like /add-gmail-agent and the skill is added. The system is infinitely customizable - you can write your own skills or have Claude build them for you.

What is the Mission Control dashboard?

Mission Control is GhostClaw's built-in dashboard accessible at localhost:3333. It provides real-time logs, task management, chat history, and research monitoring all in one place for managing your AI agent.

Can I switch AI models in GhostClaw?

Yes, you can switch between Sonnet, Opus, and Haiku models directly from Telegram by typing /model. No restart or configuration changes required.

Is GhostClaw safe to use with full system access?

GhostClaw runs on your own hardware with an MIT licence - it's yours to control. However, it has no sandboxing and gives full machine access including Bash, files, web search, and email with no permission prompts. This freedom is intentional for personal AI agents but means you're trusting your agent with complete system access. The open-source nature means you can audit the code yourself.

Categories

Use cases

Browse all AI tools on NeedAnAI