# ClawdBot is out of control Source: https://www.youtube.com/watch?v=7fltOAg8ZGI Recap page: https://rapidrecap.app/video/7fltOAg8ZGI Generated: 2026-01-27T04:04:49.493+00:00 --- ## Quick Overview The ClawdBot AI assistant, which allows users to automate tasks via messaging apps like Telegram, Discord, and WhatsApp, is open-source and self-hostable, but its security relies heavily on user configuration, as evidenced by a Shodan scan revealing many instances left exposed with zero authentication, prompting security warnings from experts. **Key Points:** - ClawdBot is an open-source, self-hosted personal AI assistant that connects to files/terminal and interacts via messaging apps like Telegram, Discord, and WhatsApp (00:20). - The creator, Peter Steinberger, uses it for tasks like managing his computer, checking emails, and even controlling his smart home devices (00:39, 02:07). - The setup process involves running an installation script and onboarding via CLI, which can be complex, especially if using Discord or WhatsApp which require developer portal setup or QR codes (02:22, 02:47). - A recent security investigation highlighted by a tweet from @fmdz387 revealed 954 public instances, with 10 requiring authentication and 429 unreachable, indicating many users exposed their bot gateways (12:53). - Another user, Alex Finn, demonstrated the power of custom skills by having his bot (Henry) use an ElevenLabs skill to call a restaurant and complete a reservation when OpenTable failed (13:31). - The creator emphasizes that the power of these agents comes with inherent risks, requiring users to be aware of security best practices, as the system can access vast amounts of personal data (03:27, 09:38, 22:58). ![Screenshot at 00:14: Andrej Karpathy's tweet detailing his rapid adoption of LLM coding capabilities, showing a shift from 80% manual coding to 80% agent coding in just two months, providing context for the cutting-edge nature of AI agents like ClawdBot.](https://ss.rapidrecap.app/screens/7fltOAg8ZGI/00-00-14.jpg) **Context:** The video features Peter Steinberger, the creator of ClawdBot, discussing his open-source, self-hosted AI assistant that integrates with various messaging platforms and system tools. The discussion covers the installation process, the agent's capabilities (like using LLMs for complex tasks and controlling devices), and recent security concerns raised on social media regarding improperly configured, publicly exposed instances of the bot. ## Detailed Analysis Peter Steinberger introduces ClawdBot as a powerful, open-source, self-hosted AI assistant that runs locally on a machine or server and interacts via messaging apps like Telegram, Discord, and WhatsApp. He details its capabilities, including managing his computer, calendar, files, health data, and even controlling his smart home lights (01:55, 02:29). He highlights the importance of the TUI (Text User Interface) for the initial setup, comparing the agent interaction style to controlling armies in old video games like Age of Empires II (04:30, 06:24). He stresses that the system is highly capable, even showing a demonstration where his bot successfully made a restaurant reservation by using an external skill (ElevenLabs) when OpenTable failed (13:31). Steinberger also addresses security concerns, referencing a Shodan scan that found 954 instances of ClawdBot exposed online, with 10 requiring auth and 429 unreachable due to users opening ports without proper security, reinforcing the warning that this powerful tool is inherently risky (12:53, 22:58). The installation guide requires Node.js, API keys for LLMs (Anthropic recommended), and setting up a messaging platform, with Telegram being the easiest, though Discord, WhatsApp, and Slack are also supported (03:41, 04:05, 22:22). ### ClawdBot Overview - Powerful, open-source, self-hosted AI assistant - Runs on local machine/server - Interacts via Telegram, Discord, WhatsApp (00:20) ### Creator Use Cases - Managing computer/files, checking emails, controlling smart home, tracking fitness goals (02:07, 03:27) ### Setup & Configuration - Install via CLI script, onboarding wizard configures Brain (LLM) and Mouth (Messaging Platform) (02:22, 02:49) ### LLM Provider Options - Anthropic (Opus/Sonnet recommended), OpenAI (GPT-4o alternative), plus others like Gemini, Google, etc. (04:06, 23:11) ### Messaging Platforms - Telegram easiest (Bot API), alternatives include Discord (Developer Portal), WhatsApp (QR code), or Slack (02:47) ### Security Concerns - Shodan scan found 954 instances, 10 exposed with auth required, 429 unreachable, highlighting user risk of improper setup (12:53, 18:56) ### Advanced Capabilities - Agents can run CLI commands, read/write files, use tools like Brave Search, and integrate with services like Twilio for voice calls (03:50, 04:54, 13:41) ![Screenshot at 00:11: The video begins with a split screen showing Peter Steinberger reacting while the main screen displays a tweet about building a billion-dollar company with no employees, setting the context of AI agent power.](https://ss.rapidrecap.app/screens/7fltOAg8ZGI/00-00-11.jpg) ![Screenshot at 00:57: The ClawdBot landing page appears, featuring the tagline 'THE AI THAT ACTUALLY DOES THINGS' and an image of its mascot \(00:59\).](https://ss.rapidrecap.app/screens/7fltOAg8ZGI/00-00-57.jpg) ![Screenshot at 01:14: A Google Trends graph showing a massive, near-vertical spike in searches for 'clawd bot' over the last 3 months, illustrating the project's rapid adoption \(01:21\).](https://ss.rapidrecap.app/screens/7fltOAg8ZGI/00-01-14.jpg) ![Screenshot at 01:58: Peter Steinberger presents at a 'Unicorn Mafia DEMO NIGHTS' event, showing the ClawdBot interface on a projector screen \(02:00\).](https://ss.rapidrecap.app/screens/7fltOAg8ZGI/00-01-58.jpg) ![Screenshot at 12:19: The presentation shifts to a command-line interface showing the successful installation and 'Doctor complete' status, indicating a working setup \(12:21\).](https://ss.rapidrecap.app/screens/7fltOAg8ZGI/00-12-19.jpg)