Today's AI Frameworks & SDKs: Fastest-Growing Projects — June 24, 2026
Today's AI Frameworks & SDKs category showcases a diverse range of projects that cater to various aspects of artificial intelligence development and deployment, from optimizing large language models for distributed systems to creating custom chips for high-performance computing. One project stands out with its innovative approach to pipeline-parallel inference across GPUs on separate machines, while others focus on integrating AI into hardware or developing full-stack agentic application platforms.
leyten/shard, with a growth score of 51.72 and 346 stars, provides a solution for pipeline-parallel LLM inference across multiple GPUs spread over different machines. This project is growing due to its unique approach in addressing the scalability challenges faced by large language models during inference, making it an attractive option for teams looking to optimize performance without compromising on hardware resources.
fguzman82/gateGPT, boasting 560 stars and a growth score of 43.96, transforms full transformers into custom chips through RTL (Register Transfer Level) design, capable of generating names on a Virtex-5 FPGA at approximately 56k tokens per second. The project's rapid growth can be attributed to its innovative approach in hardware acceleration for neural networks, offering developers and researchers an alternative path to achieving high throughput with minimal latency.
CortexPrism/cortex, with a growth score of 43.25 and 127 stars, introduces CortexPrism, an open-source agentic harness system designed to facilitate the development of intelligent applications. The project's rise in popularity is driven by its comprehensive approach to building agentic systems, which includes tools for analysis, optimization, and integration with AI agents.
Tencent-Hunyuan/UniRL, featuring 701 stars and a growth score of 38.12, presents UniRL, a framework dedicated to unified multimodal model reinforcement learning. Its strong performance in the category can be attributed to its robust architecture that supports diverse modalities and tasks, making it an essential tool for researchers and developers working on complex reinforcement learning problems.
machinefi/trio-retina, with 118 stars and a growth score of 37.72, offers a model-agnostic state layer for world models that can convert any detector into a queryable stream of events and latent states. The project's growing traction is due to its versatility in integrating various detectors and its ability to run on CPUs without requiring extensive hardware resources.
caezium/Burrow, accumulating 745 stars with a growth score of 36.77, introduces Burrow, a free and open-source macOS GUI for the Mole CLI that provides functionalities such as cleaning, uninstalling, optimizing, analyzing disk status, and monitoring live system activities. Its popularity stems from its user-friendly interface and comprehensive set of tools designed to enhance the efficiency of macOS users.
john-rocky/coreai-model-zoo, with 197 stars and a growth score of 29.86, offers a community model zoo and knowledge base for Apple Core AI models, focusing on converting Qwen3.5 and Gemma 4 end-to-end verified on-device GPUs/ANE. The project's rising interest can be attributed to its detailed documentation and practical insights into deploying advanced AI models directly on Apple devices.
OtterMind/Nubase, featuring 440 stars and a growth score of 29.59, presents Nubase, an open-source platform designed for AI coding, agentic applications, and modern product teams with integrated memory, database, storage, and authentication services. The project's growth is fueled by its comprehensive solution that simplifies the development process for AI-driven applications.
2aronS/Duel-Agents, accumulating 1,001 stars and a growth score of 26.93, offers CLI, SDK, and IDE plugins aimed at enhancing duel agents' capabilities. The project's increasing popularity is driven by its extensive range of tools that streamline the development and management processes for AI-driven systems.
VkRainB/ccMesh, with 64 stars and a growth score of 22.43, introduces ccMesh, a lightweight forwarding layer designed to intercept Claude protocol traffic and route it seamlessly between Anthropic Claude endpoints or OpenAI-compatible APIs without altering client configurations. The project's growing interest is due to its innovative approach in facilitating API compatibility and seamless integration for AI services.
These projects highlight the dynamic nature of the AI Frameworks & SDKs landscape, with each offering unique solutions that cater to different aspects of AI development and deployment.
leyten/shard, with a growth score of 51.72 and 346 stars, provides a solution for pipeline-parallel LLM inference across multiple GPUs spread over different machines. This project is growing due to its unique approach in addressing the scalability challenges faced by large language models during inference, making it an attractive option for teams looking to optimize performance without compromising on hardware resources.
fguzman82/gateGPT, boasting 560 stars and a growth score of 43.96, transforms full transformers into custom chips through RTL (Register Transfer Level) design, capable of generating names on a Virtex-5 FPGA at approximately 56k tokens per second. The project's rapid growth can be attributed to its innovative approach in hardware acceleration for neural networks, offering developers and researchers an alternative path to achieving high throughput with minimal latency.
CortexPrism/cortex, with a growth score of 43.25 and 127 stars, introduces CortexPrism, an open-source agentic harness system designed to facilitate the development of intelligent applications. The project's rise in popularity is driven by its comprehensive approach to building agentic systems, which includes tools for analysis, optimization, and integration with AI agents.
Tencent-Hunyuan/UniRL, featuring 701 stars and a growth score of 38.12, presents UniRL, a framework dedicated to unified multimodal model reinforcement learning. Its strong performance in the category can be attributed to its robust architecture that supports diverse modalities and tasks, making it an essential tool for researchers and developers working on complex reinforcement learning problems.
machinefi/trio-retina, with 118 stars and a growth score of 37.72, offers a model-agnostic state layer for world models that can convert any detector into a queryable stream of events and latent states. The project's growing traction is due to its versatility in integrating various detectors and its ability to run on CPUs without requiring extensive hardware resources.
caezium/Burrow, accumulating 745 stars with a growth score of 36.77, introduces Burrow, a free and open-source macOS GUI for the Mole CLI that provides functionalities such as cleaning, uninstalling, optimizing, analyzing disk status, and monitoring live system activities. Its popularity stems from its user-friendly interface and comprehensive set of tools designed to enhance the efficiency of macOS users.
john-rocky/coreai-model-zoo, with 197 stars and a growth score of 29.86, offers a community model zoo and knowledge base for Apple Core AI models, focusing on converting Qwen3.5 and Gemma 4 end-to-end verified on-device GPUs/ANE. The project's rising interest can be attributed to its detailed documentation and practical insights into deploying advanced AI models directly on Apple devices.
OtterMind/Nubase, featuring 440 stars and a growth score of 29.59, presents Nubase, an open-source platform designed for AI coding, agentic applications, and modern product teams with integrated memory, database, storage, and authentication services. The project's growth is fueled by its comprehensive solution that simplifies the development process for AI-driven applications.
2aronS/Duel-Agents, accumulating 1,001 stars and a growth score of 26.93, offers CLI, SDK, and IDE plugins aimed at enhancing duel agents' capabilities. The project's increasing popularity is driven by its extensive range of tools that streamline the development and management processes for AI-driven systems.
VkRainB/ccMesh, with 64 stars and a growth score of 22.43, introduces ccMesh, a lightweight forwarding layer designed to intercept Claude protocol traffic and route it seamlessly between Anthropic Claude endpoints or OpenAI-compatible APIs without altering client configurations. The project's growing interest is due to its innovative approach in facilitating API compatibility and seamless integration for AI services.
These projects highlight the dynamic nature of the AI Frameworks & SDKs landscape, with each offering unique solutions that cater to different aspects of AI development and deployment.