NemoClaw Knowledge Wiki

Gary Explains Clawdbot on Raspberry Pi

5 min read

https://www.youtube.com/watch?v=qePhufA1hSE Here is a Markdown summary of the video transcript featuring OpenClaw (formerly ClawdBot/MoltBot).

OpenClaw: The Local AI Assistant

Video Summary by Gary Explains

1. Introduction & Concept

The video introduces OpenClaw, an AI assistant described as a “blank canvas” with infinite potential. It is compared to Jarvis—a powerful virtual assistant that runs locally on your machine, operates 24/7, and can perform complex tasks autonomously.

  • Naming History: The software has undergone three name changes in three days due to trademark/similarity issues:
    • Original: ClawdBot (Too similar to Anthropic’s Claude).
    • Interim: MoltBot (Reference to lobsters shedding skin).
    • Current: OpenClaw.

2. What is OpenClaw?

OpenClaw is a gateway framework that connects several components to create an autonomous agent:

  • Local System Access: It can read/write local files and execute commands.
  • LLM Integration: Connects to models like OpenAI, Claude, or local LLMs to “understand” requests.
  • The Internet: Access to web search (via Brave Search API) and browser control.
  • Skills & Tools: It uses “function calling” to execute code, manage calendars, send emails, etc.

3. Hardware Recommendations

Gary recommends running OpenClaw on a dedicated, secondary machine to ensure it runs permanently without impacting your main workstation.

  • Gary’s Setup: Raspberry Pi 5 with an NVMe drive (fast storage, headless, low power).
  • Alternative: A Mac Mini (provides access to macOS-specific tools and integrations).

4. ⚠️ Critical Security Warning

This is a high-risk tool.

  • Full Access: The bot has read/write access to your local files and potentially your online accounts.
  • Vulnerability: It is currently susceptible to “social engineering.” For example, if it has email access, a stranger could email the bot claiming to be the owner and ask for sensitive files, and the bot might comply because it lacks robust authentication for incoming external prompts.
  • Advice: Install behind a firewall and ideally on a segregated network or secondary machine (like the Pi).

5. Installation & Configuration

  • Install: A simple one-line curl command (requires installing prerequisites like brew and Node.js).
  • LLM Setup: Users must provide API keys. Gary suggests using an OpenAI Plus subscription to use “Open Codex,” which avoids per-token API charges for heavy usage.
  • Channels: You can control the bot via Telegram, WhatsApp, Discord, etc. Gary demonstrates setting up a Telegram Bot to chat with his Raspberry Pi.
  • Personality: You can define the user (You) and the Bot (Persona). Gary named his bot “Eric” (a Monty Python reference).

6. Capabilities & Real-World Examples

The video showcases four escalating examples of what OpenClaw can do without explicit step-by-step instructions:

  1. Local File Management:
    • Prompt: “Create 10 text files in my home directory.”
    • Result: The bot immediately created the files on the Pi.
  2. Web Development & Deployment:
    • Prompt: Create a 3D rotating cube website in JS and upload it to Cloudflare Pages (Gary provided only the API token).
    • Result: The bot wrote the code, navigated the Cloudflare deployment process, handled errors (creating the project first), and returned a live URL.
  3. Software Engineering:
    • Prompt: Create a C string library, write unit tests, and upload it to a new GitLab repository.
    • Result: The bot wrote the C code, the tests, the Makefiles, and successfully pushed the project to GitLab.
  4. Autonomous Automation (The “Mind-Blowing” Example):
    • Prompt: Gary provided an API token for agentmail.to (a service he hadn’t used) and asked the bot to create an inbox and write a Python script to monitor it.
    • Result: The bot figured out the API documentation, created a custom email address, wrote a script to check that inbox every 2 minutes, and set it up to notify Gary via Telegram whenever a new email arrived.

7. Conclusion

OpenClaw brings us closer to the promise of a true “Jarvis-like” AI. While the setup requires some technical knowledge and there are significant security implications to consider, the ability of the bot to chain tasks (write code → fix errors → deploy → notify) makes it an incredibly powerful tool for automation.

Based on the video visuals and the transcript, here are the installation commands for OpenClaw. Note: Even though the name changed to OpenClaw, the URL in the video still uses the previous name “Molt.”

1. The One-Line Installer (Primary Method)

Run this command in your terminal (macOS or Linux/Raspberry Pi):

curl -fSsL https://molt.bot/install.sh | bash

2. Manual Pre-requisites (Gary’s “Pro Tip”)

In the video, Gary mentions that while the script claims to install everything, he ran into issues (specifically regarding Homebrew on the Raspberry Pi) and recommended installing some dependencies manually if the script fails or hangs. If you are on a Raspberry Pi (or Debian/Ubuntu Linux), you should ensure you have git and curl installed before running the script above:

sudo apt update
sudo apt install git curl

(Note: The script attempts to install Node.js for you, but if you have a very old version of Node, you might need to update it manually).

3. Running the Bot

Once installed, the command to start the bot and begin the onboarding process (where you enter your API keys and set up Telegram) is:

claw

⚠️ Security Reminder

As emphasized in the video, do not run this on your primary computer or a machine containing sensitive financial/personal data. This bot has full read/write access to your file system. It is highly recommended to install this on a Raspberry Pi or a dedicated secondary machine behind a firewall.

Graph View