# Origin story of OpenClaw: From 1-hour prototype to 180,000 stars of GitHub | Peter Steinberger Source: https://www.youtube.com/watch?v=0ch2xoBzf3A Recap page: https://rapidrecap.app/video/0ch2xoBzf3A Generated: 2026-02-18T05:33:28.954+00:00 --- ## Quick Overview The OpenClaw project, which achieved over 175,000 GitHub stars rapidly, originated from a one-hour prototype built by connecting WhatsApp via CLI to cloud code, demonstrating the power of simple integration and giving the creator immense fun which contrasted sharply with the subsequent high-stress, two-stage name change saga driven by crypto attackers and a polite request from Anthropic. **Key Points:** - The initial OpenClaw prototype, which felt 'really cool,' was built in one hour by hooking up WhatsApp to cloud code via CLI: 'One shot the CLI message comes in. I call the CLI with minus P. It does its magic. I get the string back and I send it back to WhatsApp.' - The project rapidly gained traction, becoming the 'fastest growing repository in GitHub history with now over 175,000 stars,' driven by the creator's desire for a personal assistant that could handle images in chat. - A mind-blowing moment occurred when the agent automatically processed an audio message by detecting the file was 'oops' (Ogg Vorbis), converting it with ffmpeg, using OpenAI for transcription, and figuring out the necessary APIs without explicit instruction. - The first name, W relay, evolved to 'Clauders' (a play on Claude with a W) and then 'Cloudbot' after a friendly request from an Anthropic employee led to immense pressure and a two-day deadline for the name change. - The transition from Cloudbot to Moldbot was disastrous because squatters stole names and accounts in seconds; the creator lost the GitHub account name, which was then used for malware serving, and the npm package was also sniped. - The final name, OpenClaw, was chosen after consulting with Sam Altman, and the creator successfully executed the rename atomically, paying $10k for a business Twitter account to claim the handle, though they were ultimately blocked from using openclaw.ai due to trademark concerns. - The creator views the viral 'Mold Book' simulation—agents talking in a Reddit-style network—as 'art' and 'the finest slop,' but notes that public reaction highlighted widespread 'AI psychosis' and gullibility regarding AI capabilities. **Context:** The interview features Peter Steinberger discussing the origin story of his highly successful open-source project, OpenClaw (previously known by names like W relay, Clauders, Cloudbot, and Moldbot). The discussion centers on the rapid development of the initial agentic system, the 'magical' emergent behavior observed, the intense real-world drama surrounding two mandatory name changes (one requested by Anthropic, the other forced by crypto squatters), and the subsequent public reaction to agentic systems exemplified by the viral 'Mold Book' demonstration. ## Detailed Analysis Peter Steinberger details how OpenClaw began as a solution to his own need for an AI personal assistant, starting with an initial one-hour prototype connecting WhatsApp through a CLI to cloud code, which felt 'really cool' because it allowed him to 'talk to my computer.' He later added image support, finding the integration seamless even on shaky mobile internet during a trip. A significant turning point was observing the agent's emergent problem-solving skills, such as automatically converting an unknown audio file format and using external APIs without being explicitly taught those steps. The project exploded in popularity, leading to a stressful name change saga; initially named Cloudbot, Steinberger received a polite but urgent request from Anthropic to change the name, which was then complicated by crypto attackers who actively monitored and snipped names and domains, causing significant operational failure during the transition to Moldbot. After almost abandoning the project due to the stress, he settled on OpenClaw after consulting with Sam Altman. Steinberger emphasizes that the project's success stems from prioritizing fun and weirdness over seriousness, noting that the agent's ability to self-introspect and modify its own TypeScript source code was a natural outcome of his debugging style. Furthermore, he addresses the security implications, cautioning against using weak local models due to increased susceptibility to prompt injection, and notes that while he made the system easy to use, users must heed security warnings about exposing backends publicly. ### Prototype Genesis - The initial concept involved creating a personal assistant AI, leading to an experiment with WhatsApp data and GPT-4; the breakthrough came in November when he built the first functional agent relay in one hour by connecting WhatsApp via CLI to cloud code. ### Emergent Capabilities - The agent displayed fascinating zero-shot abilities, including automatically detecting an audio file format, using ffmpeg for conversion, and calling the OpenAI API for transcription, which the creator found 'mind-blowing.' ### Development Philosophy - Steinberger describes his development process as playing and having fun, contrasting his current motivation with earlier projects; he views the project's success partly because 'they all take themselves too serious' while he aimed for 'weird.' ### The Naming Crisis - The project went through W relay, Clauders, Cloudbot, and Moldbot; the shift from Cloudbot to Moldbot was catastrophic due to crypto squatters successfully sniping the old account names and serving malware, leading to immense stress and near abandonment. ### Self-Modification and Open Source Impact - The agent frequently modified its own software via the agentic loop, which Steinberger used for debugging; this accessibility inspired many non-programmers to make their first pull requests, which he terms 'prompt requests,' viewing it as a win for society in creating builders. ### Security and Public Perception - Security concerns, particularly prompt injection, remain an open problem, though newer models show more resilience; Steinberger warns against using weak local models, as they are more 'gullible,' and notes that while the project is powerful, public reaction to viral demonstrations like 'Mold Book' revealed dangerous levels of 'AI psychosis' and gullibility.