OpenClaw 2.0 release announced, including shared cloud workers, persistent memory, approval controls, and experimental Swarm feature.
OpenClaw 2.0 Launches With Persistent Memory, Shared Workers, Swarm
The release adds infrastructure controls for managing agent autonomy, including approval gates and shared workers. The benchmark experiment illustrates that token costs in agents are partly a function of how much verification work they perform, not just generation.
The full picture
OpenClaw 2.0 has been released with shared cloud workers, persistent memory, widgets, dashboards, approval controls, local model support, agent interoperability, and an experimental multi-agent feature called Swarm. Chief Architect Vincent Koc was scheduled to demo the release live on YouTube on September 3. A head-to-head experiment comparing OpenClaw 2.0 against a Hermes agent on a video generation task found that differences in self-verification depth, not raw capability, drove differences in token consumption and output quality between the two agents.
How it developed
Experiment comparing OpenClaw 2.0 and Hermes agent on a video generation task published, showing verification depth drove token differences.
Sources
Want this in your inbox?
I send one email each morning with the stories that moved. If you would rather just read here, that works too.
Subscribe free