Go to TMLR homepageA Survey of the OpenClaw Ecosystem: From Platform Extensibility to Constraint Design
Abstract: Large language models have evolved into autonomous agents capable of invoking tools, using memory, and taking actions in real-world environments. Yet despite this progress, many agent systems remain difficult for ordinary users to adopt directly. OpenClaw, an open-source and self-hosted agent platform, addresses this gap through a local-first, messaging-native, and skill-extensible design that connects everyday messaging applications to LLM-powered agents. This design makes OpenClaw one of the first open personal-agent ecosystems, with ClawHub for shared Skills, Heartbeat for proactive background turns, and Moltbook as an agent-only social network. We survey this emerging ecosystem and show that the literature repeatedly highlights the same tradeoff: the openness that makes OpenClaw extensible also creates new governance, security, social, and deployment challenges. We organize the survey around four dimensions that trace this tradeoff from platform design to its ecosystem-level consequences: \emph{Platform}, where open Skills enable rapid capability growth but create new governance problems; \emph{Security}, where open Tools, Skills, Memory, and background execution expand the attack surface; \emph{Societies}, where Moltbook reveals a gap between social appearance and reliable collective intelligence; and \emph{Deployment}, where trustworthy use in robotics, healthcare, and scientific research depends on limiting agent freedom rather than expanding it. We also organize OpenClaw benchmarks into a lifecycle view of open-agent evaluation and outline future directions for treating constraints as core parts of open agent platform design. Companion repository: https://anonymous.4open.science/r/Awesome-OpenClaw-Papers/
Submission Type: Long submission (more than 12 pages of main content)
Assigned Action Editor: ~Hongyang_Zhang1
Submission Number: 9340
Loading