Skip to main content

OpenCSG Community Release Notes v2026.05

1 SkillHub as OpenClaw Registry Backend​

SkillHub now officially supports acting as a Registry Backend, enabling OpenClaw to search and install platform-hosted Skills directly — completing the end-to-end chain from skill management to Agentic execution. Instructions

  • Unified Registry: SkillHub exposes itself as a standard Skill Registry, allowing Agentic applications to fetch available Skill listings and invoke them without any additional configuration.
  • Full Toolchain Integration: The deep integration between OpenClaw and SkillHub lets developers publish, version, and invoke Skills from a single platform, significantly reducing the onboarding time for Agentic application tooling.

2 AI Gateway LLM Input/Output Logging (EE)​

AI Gateway now supports recording complete LLM request and response payloads, improving LLM observability and providing a foundation for automated training data collection.

  • Full-Chain Logging: For all LLM calls routed through AI Gateway, both the input Prompt and output Response are fully persisted, supporting issue tracing and behavior auditing.
  • Training Data Collection: Logged inputs and outputs can be exported directly in training corpus formats, providing real-world business data for continuous model iteration.

3 Serverless API Widget Parameter Configuration​

The Serverless API online Widget now supports inference parameter configuration, enabling developers to tune model settings directly in the browser and observe results in real time — greatly enhancing the interactive experience.

4 Unified Asset Creation Entry Point​

The creation entry points for models, datasets, code repositories, Spaces, and other assets are now consolidated under a single "New Resource" button, reducing navigation overhead across different pages.

5 Model and Dataset Filtering Enhancement​

Model and dataset pages now feature new filters supporting selection by parameter size range for models and by repository size range for datasets.

6 vLLM Upgraded to 0.20.1​

The vLLM inference engine has been upgraded to version 0.20.1 with support for both AMD and NVIDIA GPUs, delivering improved inference performance and broader model compatibility.

History Version​

Click Release List to view all history versions.