Today's AI Frameworks & SDKs: Fastest-Growing Projects — September 29, 2026
Today's the AI Frameworks & SDKs space, we see a continued trend of innovative projects focusing on high-performance decision models and scalable multi-agent systems. Developers are increasingly interested in frameworks that offer both local deployment options and comprehensive documentation to support broader adoption and understanding.
wfzyx/von
The von project offers an open-source System One decision model designed for sub-15ms, non-autoregressive performance as a drop-in alternative to TypeSafe Jev. With a growth score of 69.14 and over 750 stars, it stands out due to its high commit activity and the promise of local deployment capabilities.
AbdelStark/awesome-typesafe-jev
This repository serves as an extensive resource for understanding TypeSafe's System One model, including SDKs, live demos, agent tools, and independent evaluations. With a growth score of 63.96 and nearly 540 stars, it is growing due to its comprehensive documentation and community engagement.
yifanzhang-pro/recurrent-looped-tranformer
The Recurrent Looped Transformer (RLT) project provides the official codebase for RLT, which focuses on transformer-based architectures with unique recurrent looping mechanisms. Despite having fewer recent commits compared to other projects, it has gained significant traction with 905 stars and a growth score of 39.06.
Continuum-AI-Corp/OrcaBonsai-27B-Uncensored
This project offers runtime behavioral ablation for compressed LLMs, targeting the Ternary Bonsai 2 27B model without modifying or re-quantizing weights. With a growth score of 38.64 and 550 stars, it is growing due to its focus on advanced optimization techniques in large language models.
maanHimself/OpenDLSS-NR
OpenDLSS-NR provides a Vulkan reimplementation of NVIDIA's DLSS 5 Neural Rendering network, aiming for bit-exact results against the original. Despite having fewer recent commits, it has attracted 488 stars and a growth score of 36.78 due to its precise implementation and compatibility with Vulkan.
TokenRhythm/NeoHorse
NeoHorse-1 is focused on recursive self-improvement through agentic post-training with routing harness techniques. With over 1,200 stars and a growth score of 26.78, it is gaining attention for its innovative approach to self-enhancing AI agents.
Jiarui-0410/multimodal-product-discovery
This project offers a full-stack multimodal product retrieval system using CLIP, FAISS, Spring Boot, FastAPI, and PostgreSQL. Although the commit activity is low, it has garnered 125 stars and a growth score of 25.33, highlighting its potential in e-commerce applications.
AetherLabsAI/RSIAgent
RSIAgent is a training-free multi-agent framework designed for recursive self-improvement through autonomous exploration and reusable memory. It has accumulated 440 stars and a growth score of 25.09 due to its robust approach to enabling AI agents in new environments.
muellerberndt/cadence
Cadence is a machine learning library that builds consensus-style equilibrium world models continuously learning from experience. With a growth score of 23.55 and 325 stars, it stands out for its unique focus on consensus-based learning approaches.
0xBakeer/deepseek-v41-flash-spark
DeepSeek-V4.1-Flash is an ongoing project that aims to optimize deep learning workloads on a single DGX Spark using NVMe streaming and OpenAI API integration. Despite being in progress, it has gathered 120 stars and a growth score of 19.97 due to its innovative use of hardware acceleration techniques.
These projects highlight the diverse range of applications and methodologies currently explored within AI frameworks and SDKs, from decision models and transformer architectures to multi-agent systems and deep learning optimization on specialized hardware.
wfzyx/von
The von project offers an open-source System One decision model designed for sub-15ms, non-autoregressive performance as a drop-in alternative to TypeSafe Jev. With a growth score of 69.14 and over 750 stars, it stands out due to its high commit activity and the promise of local deployment capabilities.
AbdelStark/awesome-typesafe-jev
This repository serves as an extensive resource for understanding TypeSafe's System One model, including SDKs, live demos, agent tools, and independent evaluations. With a growth score of 63.96 and nearly 540 stars, it is growing due to its comprehensive documentation and community engagement.
yifanzhang-pro/recurrent-looped-tranformer
The Recurrent Looped Transformer (RLT) project provides the official codebase for RLT, which focuses on transformer-based architectures with unique recurrent looping mechanisms. Despite having fewer recent commits compared to other projects, it has gained significant traction with 905 stars and a growth score of 39.06.
Continuum-AI-Corp/OrcaBonsai-27B-Uncensored
This project offers runtime behavioral ablation for compressed LLMs, targeting the Ternary Bonsai 2 27B model without modifying or re-quantizing weights. With a growth score of 38.64 and 550 stars, it is growing due to its focus on advanced optimization techniques in large language models.
maanHimself/OpenDLSS-NR
OpenDLSS-NR provides a Vulkan reimplementation of NVIDIA's DLSS 5 Neural Rendering network, aiming for bit-exact results against the original. Despite having fewer recent commits, it has attracted 488 stars and a growth score of 36.78 due to its precise implementation and compatibility with Vulkan.
TokenRhythm/NeoHorse
NeoHorse-1 is focused on recursive self-improvement through agentic post-training with routing harness techniques. With over 1,200 stars and a growth score of 26.78, it is gaining attention for its innovative approach to self-enhancing AI agents.
Jiarui-0410/multimodal-product-discovery
This project offers a full-stack multimodal product retrieval system using CLIP, FAISS, Spring Boot, FastAPI, and PostgreSQL. Although the commit activity is low, it has garnered 125 stars and a growth score of 25.33, highlighting its potential in e-commerce applications.
AetherLabsAI/RSIAgent
RSIAgent is a training-free multi-agent framework designed for recursive self-improvement through autonomous exploration and reusable memory. It has accumulated 440 stars and a growth score of 25.09 due to its robust approach to enabling AI agents in new environments.
muellerberndt/cadence
Cadence is a machine learning library that builds consensus-style equilibrium world models continuously learning from experience. With a growth score of 23.55 and 325 stars, it stands out for its unique focus on consensus-based learning approaches.
0xBakeer/deepseek-v41-flash-spark
DeepSeek-V4.1-Flash is an ongoing project that aims to optimize deep learning workloads on a single DGX Spark using NVMe streaming and OpenAI API integration. Despite being in progress, it has gathered 120 stars and a growth score of 19.97 due to its innovative use of hardware acceleration techniques.
These projects highlight the diverse range of applications and methodologies currently explored within AI frameworks and SDKs, from decision models and transformer architectures to multi-agent systems and deep learning optimization on specialized hardware.