Today's AI Frameworks & SDKs: Fastest-Growing Projects — September 17, 2026
Today's AI Frameworks & SDKs category on pullrepo.com highlights a diverse range of projects that cater to various segments of the AI ecosystem, from multi-agent systems and model inference frameworks to lightweight development tools for MATLAB scripts and Simulink models. The top entries show a mix of high engagement metrics such as stars and commits, indicating active communities and continuous development.
AetherLabsAI/RSIAgent
This project offers a training-free multi-agent framework designed for recursive self-improvement in new environments through broad-then-deep autonomous exploration and reusable memory. With its growth score of 68.50 and over 248 stars, RSIAgent stands out due to its innovative approach to enabling agents to adapt and improve autonomously without requiring extensive training data.
0xBakeer/deepseek-v41-flash-spark
DeepSeek-V4.1-Flash is a project that runs on a single DGX Spark (GB10) and integrates resident hot experts with NVMe streaming, DSpark, and the OpenAI API. The framework's growth score of 53.50 and its high number of commits in the last month suggest significant development activity and community interest.
hkqr/my-free-code
This open-source multi-provider AI gateway provides model routing, streaming capabilities, tools for reasoning, fallbacks, and local model support, catering to Claude Code and other coding agents. With 32.81 growth score and a substantial 637 stars, my-free-code demonstrates strong community engagement and utility across various coding environments.
Zyrexnn/Cybermes
Cybermes is an autonomous offensive security framework that leverages Hermes Agent for specialized reasoning skills and multi-model LLM orchestration, aiming to enhance bug bounty and red teaming efforts. Its growth score of 32.36 and a high star count (842) reflect its relevance in the cybersecurity domain.
Nehanth/swarmllm
SwarmLLM is designed for peer-to-peer LLM inference across browser tabs, utilizing WebGPU and WebRTC to split large models over devices in real-time collaboration scenarios. With 29.78 growth score and 314 stars, this project shows promise in enabling distributed model processing and collaborative AI applications.
huawei-bayerlab/marigold-v2
Marigold V2 revisits diffusion transformers for monocular depth estimation, offering a specialized approach to visual perception tasks. Despite its niche focus, the framework has garnered 729 stars, indicating significant interest from researchers and developers in this area of computer vision.
savellonirourou107-hue/SLX-Studio
This lightweight desktop IDE supports editing and running MATLAB .m scripts and Simulink .slx models with features like visual editing, plots, parameter sweeps, and AI assistance. Its growth score of 22.37 and 116 stars suggest it is gaining traction among users looking for a more integrated development experience.
peonist-ai/halogen-flash-server
Designed to run Qwen3.8-Flash-Next on Strix Halo (gfx1151), this project aims to optimize performance through specialized hardware integration. With 20.89 growth score and 478 stars, halogen-flash-server reflects the growing interest in efficient AI model deployment across different computing platforms.
TokenRhythm/NeoHorse
NeoHorse-1 is an agentic post-training framework that aims to achieve recursive self-improvement via routing harness techniques. With a growth score of 20.65 and 462 stars, this project highlights the importance of continuous learning and improvement in AI systems.
TrenTorch/TrenTorch
Inspired by Harvard's TinyTorch, TrenTorch is designed to teach PyTorch through hands-on development experiences. Its growth score of 17.91 and 274 stars indicate that it is becoming a valuable resource for those learning and experimenting with deep learning frameworks.
These projects collectively showcase the breadth and depth of innovation in AI Frameworks & SDKs, catering to developers, researchers, and enthusiasts across various domains and use cases.
AetherLabsAI/RSIAgent
This project offers a training-free multi-agent framework designed for recursive self-improvement in new environments through broad-then-deep autonomous exploration and reusable memory. With its growth score of 68.50 and over 248 stars, RSIAgent stands out due to its innovative approach to enabling agents to adapt and improve autonomously without requiring extensive training data.
0xBakeer/deepseek-v41-flash-spark
DeepSeek-V4.1-Flash is a project that runs on a single DGX Spark (GB10) and integrates resident hot experts with NVMe streaming, DSpark, and the OpenAI API. The framework's growth score of 53.50 and its high number of commits in the last month suggest significant development activity and community interest.
hkqr/my-free-code
This open-source multi-provider AI gateway provides model routing, streaming capabilities, tools for reasoning, fallbacks, and local model support, catering to Claude Code and other coding agents. With 32.81 growth score and a substantial 637 stars, my-free-code demonstrates strong community engagement and utility across various coding environments.
Zyrexnn/Cybermes
Cybermes is an autonomous offensive security framework that leverages Hermes Agent for specialized reasoning skills and multi-model LLM orchestration, aiming to enhance bug bounty and red teaming efforts. Its growth score of 32.36 and a high star count (842) reflect its relevance in the cybersecurity domain.
Nehanth/swarmllm
SwarmLLM is designed for peer-to-peer LLM inference across browser tabs, utilizing WebGPU and WebRTC to split large models over devices in real-time collaboration scenarios. With 29.78 growth score and 314 stars, this project shows promise in enabling distributed model processing and collaborative AI applications.
huawei-bayerlab/marigold-v2
Marigold V2 revisits diffusion transformers for monocular depth estimation, offering a specialized approach to visual perception tasks. Despite its niche focus, the framework has garnered 729 stars, indicating significant interest from researchers and developers in this area of computer vision.
savellonirourou107-hue/SLX-Studio
This lightweight desktop IDE supports editing and running MATLAB .m scripts and Simulink .slx models with features like visual editing, plots, parameter sweeps, and AI assistance. Its growth score of 22.37 and 116 stars suggest it is gaining traction among users looking for a more integrated development experience.
peonist-ai/halogen-flash-server
Designed to run Qwen3.8-Flash-Next on Strix Halo (gfx1151), this project aims to optimize performance through specialized hardware integration. With 20.89 growth score and 478 stars, halogen-flash-server reflects the growing interest in efficient AI model deployment across different computing platforms.
TokenRhythm/NeoHorse
NeoHorse-1 is an agentic post-training framework that aims to achieve recursive self-improvement via routing harness techniques. With a growth score of 20.65 and 462 stars, this project highlights the importance of continuous learning and improvement in AI systems.
TrenTorch/TrenTorch
Inspired by Harvard's TinyTorch, TrenTorch is designed to teach PyTorch through hands-on development experiences. Its growth score of 17.91 and 274 stars indicate that it is becoming a valuable resource for those learning and experimenting with deep learning frameworks.
These projects collectively showcase the breadth and depth of innovation in AI Frameworks & SDKs, catering to developers, researchers, and enthusiasts across various domains and use cases.