Release Notes

2.3.0

2.3.0 introduces OpenAI-compatible Responses API support, image editing, and asynchronous generation for long-running tasks, along with full TPU accelerator support and forwarded service proxying. It also adds ZIP package downloads for code repos, mirror-backed forks, and batch repository extra info lookup, while enhancing circuit breaker reliability, resource scheduling checks, fine-tuning command validation, repository list filtering, and overall security and stability. Github

Supports cross-community federation architecture (cross-domain SSO/authentication, federated asset discovery and repository sync), compute voucher top-up and deduction settlement system, as well as Agent observability with tracing and cluster node status monitoring. (Enterprise Edition only)

2.2.0

2.2.0 strengthens AI Gateway upstream reliability and observability (with circuit breaking, fallback routing, and LLM tracing), while introducing ClawHub-compatible Agent skill publishing and version management. It also adds Argo-based Platform Dataflow workflows, administrator activity audit logs, ZIP/tar.gz code package imports, and fine-grained resource scheduling. Github

AI Gateway adds full-chain Prompt and Response audit logging for LLM requests and responses, with direct export support for model training datasets. (Enterprise Edition only)

2.1.0

2.1.0 expands AI Gateway multimodal capabilities (adding audio transcription and video generation APIs) and introduces namespace-scoped API Key management and authentication across users and organizations. It also supports LLM-powered industry tag auto-scanning, SGLang Qwen3-Guard content moderation, AMD EvalScope evaluation integration, and optimized node isolation, LFS validation, and inference architecture controls. Github

AI Gateway enhances multimodal API support (audio transcription, text/image-to-video), full compatibility with native Claude endpoints, and optimized latency and provider routing; compute resource management supports status_lost node health tracking and CPU-priority isolated scheduling. (Enterprise Edition only)

2.0.0

2.0.0 introduces repository branch management APIs and resumable LFS file uploads, while upgrading AI Gateway multimodal text-to-image capabilities and SGLang/vLLM engines (supporting Qwen 2.5, GLM 4.5, etc.). It also deepens Knative container scheduling and memory unit validation, introduces the Agent Skills architecture, JWT session middleware, and Loki-based log observability, delivering major upgrades in performance and security. Github

Launches the Compute Management Platform (cluster onboarding, HAMi/MIG virtualization slicing, global observability dashboard, and GPU-Hours billing reports), MCP Gateway unified tool governance (centralized remote MCP Server registration), and Agent Sandbox fast isolated execution runtime. (Enterprise Edition only)

1.17.0

Introduced advanced scheduling with Volcano integrated for vGPU and MIG support, enhanced infrastructure visibility with cluster resource health checks, and added deployment pending status and runtime environment tracking. The 1.17.0 update also upgrades the Agent & Skills Hub with multi-sync capabilities, improves security and authentication, and brings significant enhancements to storage and Git performance (optimized for 300k+ file repositories), reporting, system stability, and multiple bug fixes including AMD GPU inference operations. Github

Introduced Agent Memory Service for persistent context, vGPU and AMD GPU support for inference and finetuning, multi-host inference with vLLM and SGLang, PVC persistent storage for Spaces, Text-or-Image-to-Video (TI2V) multimodal capabilities, and Gradio SDK 6.2.0 upgrade, along with broad improvements to mirror syncing, content moderation, and architectural cleanup. Github

Support for local deployment of sensitive word detection services to meet enterprise compliance and data security requirements. Added License deletion capability and fixed issues related to License restrictions. Optimized JSON editing experience in the system configuration module to improve backend configuration efficiency.【EE Exclusive】

1.15.0

Introduced advanced AI Gateway capabilities (TTS and Function Calling) and significantly improved Application Spaces, community experience, resource management, and observability, with broad performance and stability enhancements. Github

Inference services now support gradual rollout of new model versions on running instances, with both Canary and Blue-Green deployment strategies available for safe and seamless upgrades. (Enterprise Edition only)

1.14.0

Migrated the core deployment scheduler to Temporal Workflow, greatly improving reliability and scalability, while expanding support for MCP, Jupyter environments, and finetuning workflows. Github

Introduce XNet Smart Trunk Accelerator, significantly enhance storage efficiency and development experience. (Enterprise Edition only)

1.12.0

Strengthened core consistency and scalability with atomic repository creation, automatic runner discovery and cluster auto-scaling, and a new streamable protocol for MCP Space execution. Github

1.11.0

Fully refactored the Runner service for secure, flexible operation both inside and outside Kubernetes, while improving error handling, i18n notifications, and workflow stability. Github

1.10.0

Introduced the DataFlow one-click data processing tool, enhanced model inference and synchronization, and delivered full server-side internationalization support. Github

1.9.0

Version 1.9.0 adds notification services, optimizes inference and multi-source synchronization functions, and supports Traditional Chinese and ultra-large file uploads. Github

A new LLM and Prompt configuration management module has been added to the admin console. (Enterprise Edition only)

1.8.0

This upgrade brings major features such as one-click deployment, performance analysis, and multi-language support, making large model management more intelligent and inference more efficient! Github

Multiple new administrative features have been introduced in the admin console, including an asset dashboard, user tagging, and user consumption reports, enabling more fine-grained asset and user management. (Enterprise Edition only)

1.7.0

Version v1.7.0 comprehensively innovates the MCP architecture, optimizes inference services, and enhances model metadata.Github

1.6.0

v1.6.0 brings several core function upgrades and performance optimizations, significantly enhancing the support for large model inference, fine-tuning, evaluation and other processes, and further improving the inference service and user interaction experience.Github

1.5.1

This version focuses on enhancing model inference, space management, UI optimization, and error handling, improving the overall user experience and platform stability. Github

1.5.0

v1.5.0 focuses on model inference framework compatibility expansion, Docker application space creation capabilities, enhanced user information management, and front-end performance optimization, providing a solid foundation for enterprises to build a more flexible and intelligent large model management platform. Github

1.4.0

v1.4.0 brings comprehensive enhancements to the tag system, system broadcasting, file list performance, and dataset preview capabilities, and further improves the background operation experience and supports mainstream large models such as Deepseek R1, providing enterprise users with a more controllable and observable large model asset management platform. Github

1.3.0

In v1.3.0, CSGHub further strengthened the tag system, inference engine compatibility and API capabilities, and significantly improved the user interaction experience and the stability and test coverage of front-end and back-end code. Github