Today's AI Frameworks & SDKs: Fastest-Growing Projects — June 18, 2026
Today's the AI Frameworks & SDKs space, there's a noticeable trend towards optimizing machine learning models for edge computing and custom hardware, alongside efforts to standardize and streamline state management across various detectors. Additionally, repositories focusing on multimodal reinforcement learning and unified model conversion continue to gain traction among developers looking to enhance their projects with advanced AI capabilities.
fguzman82/gateGPT is a project that transforms full transformers into a custom chip design using RTL (Register Transfer Level) for the microGPT model. It runs on a Virtex-5 FPGA at an impressive speed of 56,000 tokens per second. The growth score and star count suggest significant interest in this project due to its innovative approach to hardware optimization for AI models.
machinefi/trio-retina provides a model-agnostic state layer that turns any detector into a standard, queryable stream of events and latent states. This framework is particularly useful for edge computing as it runs solely on the CPU using numpy, ensuring compatibility across various devices without requiring specialized hardware. Its high growth score indicates substantial developer engagement with its unique capabilities in state management and event processing.
Tencent-Hunyuan/UniRL is a unified multimodal model reinforcement learning framework designed to facilitate research and development in AI-driven reinforcement learning applications. With over 600 stars, the project's growing popularity can be attributed to its comprehensive approach towards handling multiple modalities within reinforcement learning environments, making it an attractive resource for researchers and developers.
john-rocky/coreai-model-zoo offers a community model zoo and knowledge base dedicated to Apple Core AI models on iOS/macOS 27 devices. The repository includes verified end-to-end conversions of Qwen3.5 & Gemma 4 models, along with detailed information on conversion challenges and custom Metal kernels for optimal performance on Apple hardware. Its significant growth highlights the community's interest in optimizing AI models specifically for Apple’s ecosystem.
caezium/Burrow is a free, open-source macOS GUI for managing disk space through mole.fit, offering features like clean-up, uninstallation, optimization, and live status monitoring. The high number of stars suggests that developers are increasingly leveraging this tool to optimize their systems efficiently without relying on proprietary software solutions.
study8677/awesome-architecture is a repository containing architecture maps and tutorials aimed at helping developers think more like software architects rather than just coders. It includes templates for AI gateways, retrieval-augmented generation (RAG), agents, inference serving, and vector databases, each linked to real open-source prototypes. The project's extensive star count reflects its value in providing structured guidance on system design and architecture.
2aronS/Duel-Agents provides a suite of command-line interfaces, SDKs, and IDE plugins for working with Duel Agents, enhancing development workflows by integrating AI-driven functionalities directly into the developer’s environment. Its high popularity among developers is evident from its numerous stars, indicating active usage and contributions to the project.
huawei-csl/KVarN introduces a native vLLM KV-cache quantization backend designed to increase context capacity for agents by 3-5 times while maintaining throughput above FP16 levels and achieving FP16-level accuracy without calibration. The growing interest in this repository is reflected in its increasing star count, highlighting the demand for efficient and accurate AI model optimization tools.
RapierCraftStudios/ForgeDock transforms GitHub into a knowledge graph tailored for AI agents, offering an autonomous development pipeline that facilitates investigation, building, reviewing, merging issues, and automatically generating pull requests. The project's growing popularity is evident from its increasing star count, indicating the community’s interest in leveraging GitHub’s capabilities to streamline automated development processes.
zhnt/loushang presents an AI-native coding orchestration platform aimed at enabling unified multi-model agent runtime with stateful sessions, tool governance, and traceable delivery. Its rapid growth is likely due to the increasing demand for platforms that can effectively manage and integrate multiple AI models within a single ecosystem, providing developers with powerful tools for orchestrating complex workflows.
These projects showcase the dynamic nature of the AI frameworks landscape, highlighting innovations in hardware optimization, model conversion, reinforcement learning, system design, and agent-based development.
fguzman82/gateGPT is a project that transforms full transformers into a custom chip design using RTL (Register Transfer Level) for the microGPT model. It runs on a Virtex-5 FPGA at an impressive speed of 56,000 tokens per second. The growth score and star count suggest significant interest in this project due to its innovative approach to hardware optimization for AI models.
machinefi/trio-retina provides a model-agnostic state layer that turns any detector into a standard, queryable stream of events and latent states. This framework is particularly useful for edge computing as it runs solely on the CPU using numpy, ensuring compatibility across various devices without requiring specialized hardware. Its high growth score indicates substantial developer engagement with its unique capabilities in state management and event processing.
Tencent-Hunyuan/UniRL is a unified multimodal model reinforcement learning framework designed to facilitate research and development in AI-driven reinforcement learning applications. With over 600 stars, the project's growing popularity can be attributed to its comprehensive approach towards handling multiple modalities within reinforcement learning environments, making it an attractive resource for researchers and developers.
john-rocky/coreai-model-zoo offers a community model zoo and knowledge base dedicated to Apple Core AI models on iOS/macOS 27 devices. The repository includes verified end-to-end conversions of Qwen3.5 & Gemma 4 models, along with detailed information on conversion challenges and custom Metal kernels for optimal performance on Apple hardware. Its significant growth highlights the community's interest in optimizing AI models specifically for Apple’s ecosystem.
caezium/Burrow is a free, open-source macOS GUI for managing disk space through mole.fit, offering features like clean-up, uninstallation, optimization, and live status monitoring. The high number of stars suggests that developers are increasingly leveraging this tool to optimize their systems efficiently without relying on proprietary software solutions.
study8677/awesome-architecture is a repository containing architecture maps and tutorials aimed at helping developers think more like software architects rather than just coders. It includes templates for AI gateways, retrieval-augmented generation (RAG), agents, inference serving, and vector databases, each linked to real open-source prototypes. The project's extensive star count reflects its value in providing structured guidance on system design and architecture.
2aronS/Duel-Agents provides a suite of command-line interfaces, SDKs, and IDE plugins for working with Duel Agents, enhancing development workflows by integrating AI-driven functionalities directly into the developer’s environment. Its high popularity among developers is evident from its numerous stars, indicating active usage and contributions to the project.
huawei-csl/KVarN introduces a native vLLM KV-cache quantization backend designed to increase context capacity for agents by 3-5 times while maintaining throughput above FP16 levels and achieving FP16-level accuracy without calibration. The growing interest in this repository is reflected in its increasing star count, highlighting the demand for efficient and accurate AI model optimization tools.
RapierCraftStudios/ForgeDock transforms GitHub into a knowledge graph tailored for AI agents, offering an autonomous development pipeline that facilitates investigation, building, reviewing, merging issues, and automatically generating pull requests. The project's growing popularity is evident from its increasing star count, indicating the community’s interest in leveraging GitHub’s capabilities to streamline automated development processes.
zhnt/loushang presents an AI-native coding orchestration platform aimed at enabling unified multi-model agent runtime with stateful sessions, tool governance, and traceable delivery. Its rapid growth is likely due to the increasing demand for platforms that can effectively manage and integrate multiple AI models within a single ecosystem, providing developers with powerful tools for orchestrating complex workflows.
These projects showcase the dynamic nature of the AI frameworks landscape, highlighting innovations in hardware optimization, model conversion, reinforcement learning, system design, and agent-based development.