Today's AI Frameworks & SDKs: Fastest-Growing Projects — September 30, 2026
This week, the AI Frameworks & SDKs space continues to thrive with a variety of projects that cater to different needs, from decision-making models to CAD integration and machine learning libraries. Among these tools, wfzyx/von stands out for its high growth score, showcasing the ongoing interest in efficient and non-autoregressive decision models.
wfzyx/von
The open-source System One decision model is designed to provide sub-15ms performance as a local drop-in alternative to TypeSafe Jev. With a robust 64.33 growth score and over 773 stars, this project captures attention for its speed and efficiency in decision-making processes.
CaptureGrubEnchant/SolidWorks
This repository connects an AI assistant directly to running SolidWorks instances, enabling features like sketching, extruding, filleting, and exporting STEP/STL files. With a growth score of 63.00 and 372 stars, it highlights the increasing demand for AI integration in CAD software.
AbdelStark/awesome-typesafe-jev
A comprehensive guide to TypeSafe's System One model, this repository includes SDKs, live demos, and independent evaluations. Its high growth score of 59.58 and over 544 stars reflect the community's interest in detailed resources for understanding and leveraging advanced decision models.
yifanzhang-pro/recurrent-looped-tranformer
The Recurrent Looped Transformer (RLT) project focuses on developing a new type of transformer architecture, aiming to enhance performance through recurrent loops. With 905 stars and a growth score of 36.81, it demonstrates ongoing interest in innovative transformer designs.
Continuum-AI-Corp/OrcaBonsai-27B-Uncensored
This project offers runtime behavioral ablation for compressed large language models (LLMs), targeting the Ternary Bonsai 2 27B model. Its growth score of 35.71 and 554 stars indicate a strong community engagement with optimization techniques for LLMs.
maanHimself/OpenDLSS-NR
A Vulkan reimplementation of NVIDIA's DLSS 5 Neural Rendering network, this project aims to be bit-exact against the original implementation. With 34.65 growth and 510 stars, it highlights the importance of open-source alternatives for graphics processing.
TokenRhythm/NeoHorse
Focusing on recursive self-improvement via agentic post-training with routing harness, this project aims to enhance decision-making capabilities through iterative improvements. Its growth score of 27.04 and 1,280 stars suggest a growing interest in self-improving AI models.
AetherLabsAI/RSIAgent
This training-free multi-agent framework supports recursive self-improvement in new environments through autonomous exploration and reusable memory. With a growth score of 23.68 and 442 stars, it indicates the increasing focus on developing intelligent systems that learn from experience without extensive training.
muellerberndt/cadence
A machine learning library for building consensus-style equilibrium world models that continuously learn from experience, this project is gaining traction with a growth score of 22.59 and over 328 stars, reflecting the interest in continuous learning frameworks.
mizorewww/laya-mlx
Providing a native MLX runtime for Laya typed decision models, this tool supports short decisions on M3 Max without requiring text generation or cloud APIs. Its growth score of 21.35 and an impressive 6,631 stars underscore the community's interest in efficient, local AI solutions.
Today's report highlights a diverse array of projects across various domains within AI frameworks and SDKs, reflecting both established interests and emerging trends in the field.
wfzyx/von
The open-source System One decision model is designed to provide sub-15ms performance as a local drop-in alternative to TypeSafe Jev. With a robust 64.33 growth score and over 773 stars, this project captures attention for its speed and efficiency in decision-making processes.
CaptureGrubEnchant/SolidWorks
This repository connects an AI assistant directly to running SolidWorks instances, enabling features like sketching, extruding, filleting, and exporting STEP/STL files. With a growth score of 63.00 and 372 stars, it highlights the increasing demand for AI integration in CAD software.
AbdelStark/awesome-typesafe-jev
A comprehensive guide to TypeSafe's System One model, this repository includes SDKs, live demos, and independent evaluations. Its high growth score of 59.58 and over 544 stars reflect the community's interest in detailed resources for understanding and leveraging advanced decision models.
yifanzhang-pro/recurrent-looped-tranformer
The Recurrent Looped Transformer (RLT) project focuses on developing a new type of transformer architecture, aiming to enhance performance through recurrent loops. With 905 stars and a growth score of 36.81, it demonstrates ongoing interest in innovative transformer designs.
Continuum-AI-Corp/OrcaBonsai-27B-Uncensored
This project offers runtime behavioral ablation for compressed large language models (LLMs), targeting the Ternary Bonsai 2 27B model. Its growth score of 35.71 and 554 stars indicate a strong community engagement with optimization techniques for LLMs.
maanHimself/OpenDLSS-NR
A Vulkan reimplementation of NVIDIA's DLSS 5 Neural Rendering network, this project aims to be bit-exact against the original implementation. With 34.65 growth and 510 stars, it highlights the importance of open-source alternatives for graphics processing.
TokenRhythm/NeoHorse
Focusing on recursive self-improvement via agentic post-training with routing harness, this project aims to enhance decision-making capabilities through iterative improvements. Its growth score of 27.04 and 1,280 stars suggest a growing interest in self-improving AI models.
AetherLabsAI/RSIAgent
This training-free multi-agent framework supports recursive self-improvement in new environments through autonomous exploration and reusable memory. With a growth score of 23.68 and 442 stars, it indicates the increasing focus on developing intelligent systems that learn from experience without extensive training.
muellerberndt/cadence
A machine learning library for building consensus-style equilibrium world models that continuously learn from experience, this project is gaining traction with a growth score of 22.59 and over 328 stars, reflecting the interest in continuous learning frameworks.
mizorewww/laya-mlx
Providing a native MLX runtime for Laya typed decision models, this tool supports short decisions on M3 Max without requiring text generation or cloud APIs. Its growth score of 21.35 and an impressive 6,631 stars underscore the community's interest in efficient, local AI solutions.
Today's report highlights a diverse array of projects across various domains within AI frameworks and SDKs, reflecting both established interests and emerging trends in the field.