The open source AI assistant project now known as OpenClaw is once again redefining itself, both in name and in scope. Previously called Clawdbot, and briefly Moltbot following a legal challenge from Anthropic over naming conflicts, the project has settled on OpenClaw after additional trademark research and direct permission checks. Its creator, Austrian developer Peter Steinberger, has framed the change as a practical reset rather than a reaction, signaling a desire to stabilize the project’s identity as its community rapidly expands.
Despite its youth, OpenClaw has drawn substantial attention from developers. In just two months, the project accumulated more than 100,000 GitHub stars, reflecting strong interest in personal, locally run AI agents that can operate inside existing chat platforms. Steinberger has been clear that OpenClaw is no longer something he can manage alone, and he has begun adding maintainers from the open source community to help guide development.
What has pushed OpenClaw beyond a typical developer tool is the emergence of community-driven offshoots. One of the most notable is Moltbook, an experimental social network where AI assistants interact with one another rather than serving human users directly. On the platform, AI agents exchange information, discuss technical topics, and even share methods for coordinating behavior. The system relies on downloadable instruction files, or skills, that teach OpenClaw agents how to participate in these discussions and periodically check for updates.
The idea has attracted commentary from prominent figures in the AI community. Andrej Karpathy described the phenomenon as unusually close to science fiction, highlighting how user-created agents appear to be self-organizing. British programmer Simon Willison has also pointed to Moltbook as an unusually active experiment, while warning that agents designed to fetch instructions from the internet raise clear security concerns.
Security remains the project’s most consistent caveat. OpenClaw is designed to run on users’ own machines and connect to chat apps they already use, which amplifies the risks if misconfigured. Steinberger has acknowledged that issues like prompt injection remain unsolved across the industry, and that OpenClaw should not be deployed casually or granted access to primary communication accounts. Project maintainers have echoed this stance, emphasizing that the tool is currently suited for experienced users who understand command-line environments and security trade-offs.
To support ongoing development, OpenClaw has begun accepting sponsorships, though Steinberger says funds are earmarked for compensating maintainers rather than personal income. The project’s backers include several well-known founders and developers, underscoring confidence in its long-term potential, even if mainstream adoption is still distant.
For now, OpenClaw sits in a transitional space: technically ambitious, culturally intriguing, and openly experimental. Its growing ecosystem, including AI-to-AI social spaces, offers a glimpse into how agentic AI might evolve, while also reinforcing how much work remains before such systems are safe or accessible for general use.
