AI Frameworks & SDKs: Fastest-Growing Projects — May 05, 2026
Today's the AI Frameworks & SDKs space, we're seeing a surge in tools that cater to specific use cases, such as AI-assisted trading and on-device AI for Android. Additionally, there's a growing interest in frameworks that enable efficient and scalable deployment of large language models (LLMs). Our top 10 list reflects this trend, with repositories showcasing innovative approaches to serving LLMs, autonomous penetration testing, and vulnerability research.
HKUDS-AI/polymarket-ai-trading is a standout example, boasting an impressive Growth Score of 99.31 and 140 stars. This repository offers AI-assisted paper trading for Polymarket, leveraging OpenAI market insights, vector search, and Kelly sizing to inform trading decisions. Its rapid growth can be attributed to the increasing popularity of AI-driven trading strategies.
noonghunna/club-3090 has garnered significant attention with a Growth Score of 83.50 and 488 stars. This community-driven repository provides recipes for serving LLMs on RTX 3090, supporting multiple engines like vLLM and llama.cpp. Its growth is likely driven by the demand for efficient deployment of LLMs on high-performance hardware.
1bananachicken/MaaNTE has gained traction with a Growth Score of 81.50 and 669 stars. Although its description is in Chinese, it appears to be an AI-powered assistant built using the MAAF framework. Its popularity may stem from its effectiveness as a practical tool for users.
alash3al/stash boasts a respectable Growth Score of 59.55 and 644 stars. This repository introduces Stash, a persistent memory layer for AI agents that stores episodes, facts, and working context in Postgres. Its growth can be attributed to the need for efficient data management in AI applications.
esengine/reasonix has attracted attention with a Growth Score of 35.43 and 329 stars. This DeepSeek-native agent framework offers features like Cache-First Loop, R1 Thought Harvesting, and Tool-Call Repair. Its growth is likely driven by its innovative approach to agent-based AI development.
oritera/Cairn has gained interest with a Growth Score of 31.88 and 654 stars. This general-purpose state-space search engine was initially validated on autonomous penetration testing. Its popularity may stem from its versatility in tackling complex problems.
jegly/Box has garnered attention with a Growth Score of 20.39 and 290 stars. This private on-device AI suite for Android offers features like llama.cpp, whisper.cpp, and stable-diffusion.cpp, as well as NPU/TPU/GPU acceleration. Its growth can be attributed to the demand for secure and efficient on-device AI capabilities.
unclebob/swarm-forge has a Growth Score of 19.58 and 441 stars. This simple tool coordinates several AI agents, making it an attractive solution for developers working with multi-agent systems. Its popularity may stem from its ease of use and effectiveness in streamlining agent coordination.
FonaTech/Project_Chronos boasts a Growth Score of 19.56 and 173 stars. This project introduces zero-stall MoE inference via lookahead prediction and async DMA prefetching, optimized for SSD I/O with hybrid MLA+sliding window attention. Its growth can be attributed to the increasing interest in efficient inference techniques.
416rehman/DeepZero has gained attention with a Growth Score of 18.02 and 388 stars. This automated vulnerability research framework parses, decompiles, and analyzes thousands of Windows kernel drivers for exploitable IOCTLs using AI agents. Its popularity may stem from its innovative approach to vulnerability research.
HKUDS-AI/polymarket-ai-trading is a standout example, boasting an impressive Growth Score of 99.31 and 140 stars. This repository offers AI-assisted paper trading for Polymarket, leveraging OpenAI market insights, vector search, and Kelly sizing to inform trading decisions. Its rapid growth can be attributed to the increasing popularity of AI-driven trading strategies.
noonghunna/club-3090 has garnered significant attention with a Growth Score of 83.50 and 488 stars. This community-driven repository provides recipes for serving LLMs on RTX 3090, supporting multiple engines like vLLM and llama.cpp. Its growth is likely driven by the demand for efficient deployment of LLMs on high-performance hardware.
1bananachicken/MaaNTE has gained traction with a Growth Score of 81.50 and 669 stars. Although its description is in Chinese, it appears to be an AI-powered assistant built using the MAAF framework. Its popularity may stem from its effectiveness as a practical tool for users.
alash3al/stash boasts a respectable Growth Score of 59.55 and 644 stars. This repository introduces Stash, a persistent memory layer for AI agents that stores episodes, facts, and working context in Postgres. Its growth can be attributed to the need for efficient data management in AI applications.
esengine/reasonix has attracted attention with a Growth Score of 35.43 and 329 stars. This DeepSeek-native agent framework offers features like Cache-First Loop, R1 Thought Harvesting, and Tool-Call Repair. Its growth is likely driven by its innovative approach to agent-based AI development.
oritera/Cairn has gained interest with a Growth Score of 31.88 and 654 stars. This general-purpose state-space search engine was initially validated on autonomous penetration testing. Its popularity may stem from its versatility in tackling complex problems.
jegly/Box has garnered attention with a Growth Score of 20.39 and 290 stars. This private on-device AI suite for Android offers features like llama.cpp, whisper.cpp, and stable-diffusion.cpp, as well as NPU/TPU/GPU acceleration. Its growth can be attributed to the demand for secure and efficient on-device AI capabilities.
unclebob/swarm-forge has a Growth Score of 19.58 and 441 stars. This simple tool coordinates several AI agents, making it an attractive solution for developers working with multi-agent systems. Its popularity may stem from its ease of use and effectiveness in streamlining agent coordination.
FonaTech/Project_Chronos boasts a Growth Score of 19.56 and 173 stars. This project introduces zero-stall MoE inference via lookahead prediction and async DMA prefetching, optimized for SSD I/O with hybrid MLA+sliding window attention. Its growth can be attributed to the increasing interest in efficient inference techniques.
416rehman/DeepZero has gained attention with a Growth Score of 18.02 and 388 stars. This automated vulnerability research framework parses, decompiles, and analyzes thousands of Windows kernel drivers for exploitable IOCTLs using AI agents. Its popularity may stem from its innovative approach to vulnerability research.