Today's AI Frameworks & SDKs: Fastest-Growing Projects — September 26, 2026
This week, the AI Frameworks & SDKs space continues to show significant momentum with several projects gaining traction across various subdomains of artificial intelligence and machine learning. Notable among them are innovative decision models and multi-agent systems that cater to diverse applications ranging from model optimization to autonomous exploration.
wfzyx/von is an open-source System One decision model designed for quick execution, aiming to provide a non-autoregressive local alternative to TypeSafe Jev with sub-15ms performance. With its high growth score of 89.44 and over 678 stars on GitHub, this project stands out due to its promising performance metrics and active development cycle.
AbdelStark/awesome-typesafe-jev serves as a comprehensive resource for TypeSafe's System One model, offering SDKs, live demos, agent tools, and independent evaluations. This repository has garnered 81.83 in growth score and 519 stars, reflecting its value as an educational and practical guide within the AI community.
Continuum-AI-Corp/OrcaBonsai-27B-Uncensored focuses on runtime behavioral ablation for compressed large language models (LLMs), specifically targeting Ternary Bonsai 2 with a 27-billion parameter model. With a growth score of 52.25 and 539 stars, this project is growing due to its targeted approach in optimizing LLMs without modifying weights or re-quantizing them.
yifanzhang-pro/recurrent-looped-tranformer introduces the Recurrent Looped Transformer (RLT), a framework designed for specific applications within machine learning. This repository has accumulated 47.46 in growth score and 906 stars, indicating strong interest from developers and researchers looking to explore new transformer architectures.
maanHimself/OpenDLSS-NR offers a Vulkan reimplementation of NVIDIA's DLSS 5 Neural Rendering network, ensuring bit-exact replication against the original implementation. With 46.33 in growth score and 397 stars, this project is notable for its commitment to precision and compatibility with high-performance graphics APIs.
AetherLabsAI/RSIAgent presents a training-free multi-agent framework aimed at recursive self-improvement through autonomous exploration and reusable memory techniques. Its growth score of 30.19 and 422 stars suggest that it is gaining traction among researchers interested in developing efficient, exploratory AI systems.
muellerberndt/cadence is a machine learning library focused on building consensus-style equilibrium world models that continuously learn from experience. With a growth score of 26.66 and 308 stars, this project highlights the importance of consensus-based learning in creating adaptive AI systems.
TokenRhythm/NeoHorse features NeoHorse-1, an approach towards recursive self-improvement via agentic post-training with routing harness mechanisms. This repository has seen a growth score of 25.14 and attracted 1,001 stars, underscoring its significance in advancing the capabilities of AI agents.
0xBakeer/deepseek-v41-flash-spark showcases DeepSeek-V4.1-Flash running on a single DGX Spark with NVMe streaming and OpenAI API integration. With a growth score of 23.62 and 120 stars, this project is notable for its optimization efforts in high-performance computing environments.
Nehanth/swarmllm enables peer-to-peer LLM inference across browser tabs using WebGPU and WebRTC, splitting a large model (e.g., 27B) across devices within a room. This innovative approach has earned it a growth score of 23.10 and 450 stars, highlighting its potential in distributed computing scenarios for AI models.
These projects collectively demonstrate the dynamic nature of the AI Frameworks & SDKs space, with developers actively pushing boundaries in areas such as decision modeling, model optimization, multi-agent systems, and distributed inference.
wfzyx/von is an open-source System One decision model designed for quick execution, aiming to provide a non-autoregressive local alternative to TypeSafe Jev with sub-15ms performance. With its high growth score of 89.44 and over 678 stars on GitHub, this project stands out due to its promising performance metrics and active development cycle.
AbdelStark/awesome-typesafe-jev serves as a comprehensive resource for TypeSafe's System One model, offering SDKs, live demos, agent tools, and independent evaluations. This repository has garnered 81.83 in growth score and 519 stars, reflecting its value as an educational and practical guide within the AI community.
Continuum-AI-Corp/OrcaBonsai-27B-Uncensored focuses on runtime behavioral ablation for compressed large language models (LLMs), specifically targeting Ternary Bonsai 2 with a 27-billion parameter model. With a growth score of 52.25 and 539 stars, this project is growing due to its targeted approach in optimizing LLMs without modifying weights or re-quantizing them.
yifanzhang-pro/recurrent-looped-tranformer introduces the Recurrent Looped Transformer (RLT), a framework designed for specific applications within machine learning. This repository has accumulated 47.46 in growth score and 906 stars, indicating strong interest from developers and researchers looking to explore new transformer architectures.
maanHimself/OpenDLSS-NR offers a Vulkan reimplementation of NVIDIA's DLSS 5 Neural Rendering network, ensuring bit-exact replication against the original implementation. With 46.33 in growth score and 397 stars, this project is notable for its commitment to precision and compatibility with high-performance graphics APIs.
AetherLabsAI/RSIAgent presents a training-free multi-agent framework aimed at recursive self-improvement through autonomous exploration and reusable memory techniques. Its growth score of 30.19 and 422 stars suggest that it is gaining traction among researchers interested in developing efficient, exploratory AI systems.
muellerberndt/cadence is a machine learning library focused on building consensus-style equilibrium world models that continuously learn from experience. With a growth score of 26.66 and 308 stars, this project highlights the importance of consensus-based learning in creating adaptive AI systems.
TokenRhythm/NeoHorse features NeoHorse-1, an approach towards recursive self-improvement via agentic post-training with routing harness mechanisms. This repository has seen a growth score of 25.14 and attracted 1,001 stars, underscoring its significance in advancing the capabilities of AI agents.
0xBakeer/deepseek-v41-flash-spark showcases DeepSeek-V4.1-Flash running on a single DGX Spark with NVMe streaming and OpenAI API integration. With a growth score of 23.62 and 120 stars, this project is notable for its optimization efforts in high-performance computing environments.
Nehanth/swarmllm enables peer-to-peer LLM inference across browser tabs using WebGPU and WebRTC, splitting a large model (e.g., 27B) across devices within a room. This innovative approach has earned it a growth score of 23.10 and 450 stars, highlighting its potential in distributed computing scenarios for AI models.
These projects collectively demonstrate the dynamic nature of the AI Frameworks & SDKs space, with developers actively pushing boundaries in areas such as decision modeling, model optimization, multi-agent systems, and distributed inference.