Today's AI Frameworks & SDKs: Fastest-Growing Projects — June 10, 2026
This week, the AI Frameworks & SDKs space continues to evolve rapidly with a strong emphasis on multimodal learning and automation. Developers are showing particular interest in tools that streamline model training and annotation processes while also providing robust support for agent-based systems. One standout is Tencent-Hunyuan/UniRL, which leads Today's list with an impressive growth score.
Tencent-Hunyuan/UniRL is a framework designed to facilitate unified multimodal model reinforcement learning. With 87.00 in growth score and 267 stars, it has gained significant traction due to its unique approach to integrating various modalities for reinforcement learning tasks.
2aronS/Duel-Agents offers a comprehensive suite of tools including CLI, SDK, and IDE plugins for developing Duel Agents. This repository has seen substantial interest with 54.77 in growth score and over 1,000 stars, likely due to its wide array of features and ease of use across different development environments.
study8677/awesome-architecture provides a collection of architecture maps for software systems that incorporate AI components such as gateways, retrieval agents, and inference serving. The project's bilingual content and detailed tutorials have attracted 1,233 stars and a growth score of 54.00, reflecting its broad appeal to developers working on complex AI-integrated projects.
Somnusochi/VLM-AutoYOLO is an end-to-end platform for object detection that includes automatic annotation using vision-language models like NVIDIA LocateAnything-3B and YOLO training pipelines. With a growth score of 45.88 and 104 stars, its innovative approach to automating the labeling process stands out.
caezium/Burrow is a macOS GUI for managing system tasks such as cleaning and optimizing disk space through a command-line interface called Mole CLI (mo). The project's unique feature set and user-friendly interface have contributed to its growth score of 45.80 and 253 stars.
huawei-csl/KVarN provides a native vLLM KV-cache quantization backend that enhances the performance of agents by offering increased context and throughput while maintaining accuracy at an FP16 level. With a growth score of 41.29 and 385 stars, KVarN's ability to significantly improve agent capabilities without complex calibration has attracted considerable attention.
simonlin1212/TradingAgents-astock introduces a multi-agent investment research framework tailored for the A-share market in China, complete with AI analysts simulating real-world trading scenarios. Its 39.89 growth score and 1,097 stars highlight its relevance to financial markets where sophisticated analysis is crucial.
zhnt/loushang presents an AI-native coding orchestration platform that supports stateful sessions for multi-model agents and includes features like tool governance and traceable delivery. With a growth score of 34.17 and 118 stars, it addresses the growing need for efficient management in complex AI environments.
modelstudioai/cli is an official CLI built to facilitate model exposure, search, multimodal capabilities, and workflow management within Model Studio. Its structured tool calls have garnered a growth score of 32.65 and 216 stars, indicating its utility in simplifying the development process for AI frameworks.
Finally, huggingface/cadgenbench offers a benchmark suite specifically designed to evaluate AI-driven CAD generation and editing systems. With a modest growth score of 23.36 but still accumulating 48 stars, it provides valuable metrics for developers working on generative design applications in the CAD space.
Tencent-Hunyuan/UniRL is a framework designed to facilitate unified multimodal model reinforcement learning. With 87.00 in growth score and 267 stars, it has gained significant traction due to its unique approach to integrating various modalities for reinforcement learning tasks.
2aronS/Duel-Agents offers a comprehensive suite of tools including CLI, SDK, and IDE plugins for developing Duel Agents. This repository has seen substantial interest with 54.77 in growth score and over 1,000 stars, likely due to its wide array of features and ease of use across different development environments.
study8677/awesome-architecture provides a collection of architecture maps for software systems that incorporate AI components such as gateways, retrieval agents, and inference serving. The project's bilingual content and detailed tutorials have attracted 1,233 stars and a growth score of 54.00, reflecting its broad appeal to developers working on complex AI-integrated projects.
Somnusochi/VLM-AutoYOLO is an end-to-end platform for object detection that includes automatic annotation using vision-language models like NVIDIA LocateAnything-3B and YOLO training pipelines. With a growth score of 45.88 and 104 stars, its innovative approach to automating the labeling process stands out.
caezium/Burrow is a macOS GUI for managing system tasks such as cleaning and optimizing disk space through a command-line interface called Mole CLI (mo). The project's unique feature set and user-friendly interface have contributed to its growth score of 45.80 and 253 stars.
huawei-csl/KVarN provides a native vLLM KV-cache quantization backend that enhances the performance of agents by offering increased context and throughput while maintaining accuracy at an FP16 level. With a growth score of 41.29 and 385 stars, KVarN's ability to significantly improve agent capabilities without complex calibration has attracted considerable attention.
simonlin1212/TradingAgents-astock introduces a multi-agent investment research framework tailored for the A-share market in China, complete with AI analysts simulating real-world trading scenarios. Its 39.89 growth score and 1,097 stars highlight its relevance to financial markets where sophisticated analysis is crucial.
zhnt/loushang presents an AI-native coding orchestration platform that supports stateful sessions for multi-model agents and includes features like tool governance and traceable delivery. With a growth score of 34.17 and 118 stars, it addresses the growing need for efficient management in complex AI environments.
modelstudioai/cli is an official CLI built to facilitate model exposure, search, multimodal capabilities, and workflow management within Model Studio. Its structured tool calls have garnered a growth score of 32.65 and 216 stars, indicating its utility in simplifying the development process for AI frameworks.
Finally, huggingface/cadgenbench offers a benchmark suite specifically designed to evaluate AI-driven CAD generation and editing systems. With a modest growth score of 23.36 but still accumulating 48 stars, it provides valuable metrics for developers working on generative design applications in the CAD space.