Today's AI Frameworks & SDKs: Fastest-Growing Projects — May 23, 2026
Today's the AI Frameworks & SDKs space, we continue to see a strong focus on high-performance inference engines and multi-agent systems designed for specific use cases such as trading and automation tasks. The projects highlighted this week span from innovative approaches to speed up language model inference to sophisticated frameworks that manage agent interactions and data sources tailored for A-share markets.
lightseekorg/tokenspeed is a speed-of-light LLM inference engine, aiming to significantly reduce the latency in processing large language models. With its impressive 58.53 growth score and over 1,000 stars on GitHub, it stands out as one of Today's top performers, likely due to its high performance focus and active development.
simonlin1212/TradingAgents-astock offers a multi-agent investment research framework specifically designed for the A-share market in China. The project features seven AI analysts engaged in debates based on specific trading rules and risk assessments, making it an interesting tool for those interested in algorithmic trading tailored to Chinese financial markets. Its 55.15 growth score indicates significant interest from developers looking to leverage multi-agent systems in finance.
1bananachicken/MaaNTE, with its unique name and description suggesting it's driven by the MAAFramework, appears to offer an automation assistant for a game or application environment. Despite its playful title, it has garnered 1,463 stars on GitHub, reflecting strong community interest in the framework’s capabilities beyond just gaming applications.
noonghunna/club-3090 provides recipes and configurations for serving large language models (LLMs) on RTX 3090 GPUs. The project is model-agnostic and supports multiple engines like vLLM, llama.cpp, and SGLang, making it a versatile choice for developers looking to optimize their LLMs on specific hardware setups. Its growth score of 35.84, along with 1,036 stars, underscores its relevance in the growing ecosystem of AI model deployment.
jhaizhou-ops/pinrule introduces a universal framework for setting behavioral rules for AI systems to prevent drift in long-term tasks. The project is designed to work without requiring any large language models or network connectivity and includes preset scenarios that can be switched with just one line of code, making it highly flexible for various use cases. Its steady growth score of 31.60 and modest but growing number of stars suggest its utility in ensuring consistent AI behavior across different applications.
Ontos-AI/knowhere is a tool designed to extract, parse, and output structured data chunks that are ready for integration with AI agents or retrieval-augmented generation (RAG) systems. With 406 stars on GitHub, it reflects the growing demand for tools that facilitate the processing of complex datasets into easily consumable formats for AI applications.
alash3al/stash is a persistent memory layer designed to provide working context and episodes storage for AI agents using Postgres. The project includes an MCP server and can be self-hosted with minimal infrastructure, aligning well with trends towards on-premises AI solutions. With 699 stars, it indicates strong community interest in tools that enhance the persistence and state management capabilities of AI systems.
enmanuelmag/agent-harness-kit offers a scaffolding kit for running structured multi-agent workflows in various codebases without being tied to specific providers or platforms. This flexibility is highlighted by its 158 stars, suggesting developers are looking for modular tools that can be easily integrated into existing systems.
stoaaadev/stoa presents a framework for managing multi-agent swarms with autonomous agents and shared mesh networking, enabling complex interactions in decentralized environments. With 26 stars, it reflects niche interest from developers working on advanced swarm intelligence applications where such frameworks are crucial.
Lastly, TheRunicDev/MaaNTE appears to be another iteration of an automation assistant framework, possibly with a focus on game mechanics and user convenience features like auto-fishing or story dialogue skipping. Its 20.27 growth score and 431 stars indicate steady interest in automation tools that enhance gaming experiences through AI.
These projects reflect the diversity and innovation within the AI frameworks & SDKs space, from high-performance computing to multi-agent systems and beyond.
lightseekorg/tokenspeed is a speed-of-light LLM inference engine, aiming to significantly reduce the latency in processing large language models. With its impressive 58.53 growth score and over 1,000 stars on GitHub, it stands out as one of Today's top performers, likely due to its high performance focus and active development.
simonlin1212/TradingAgents-astock offers a multi-agent investment research framework specifically designed for the A-share market in China. The project features seven AI analysts engaged in debates based on specific trading rules and risk assessments, making it an interesting tool for those interested in algorithmic trading tailored to Chinese financial markets. Its 55.15 growth score indicates significant interest from developers looking to leverage multi-agent systems in finance.
1bananachicken/MaaNTE, with its unique name and description suggesting it's driven by the MAAFramework, appears to offer an automation assistant for a game or application environment. Despite its playful title, it has garnered 1,463 stars on GitHub, reflecting strong community interest in the framework’s capabilities beyond just gaming applications.
noonghunna/club-3090 provides recipes and configurations for serving large language models (LLMs) on RTX 3090 GPUs. The project is model-agnostic and supports multiple engines like vLLM, llama.cpp, and SGLang, making it a versatile choice for developers looking to optimize their LLMs on specific hardware setups. Its growth score of 35.84, along with 1,036 stars, underscores its relevance in the growing ecosystem of AI model deployment.
jhaizhou-ops/pinrule introduces a universal framework for setting behavioral rules for AI systems to prevent drift in long-term tasks. The project is designed to work without requiring any large language models or network connectivity and includes preset scenarios that can be switched with just one line of code, making it highly flexible for various use cases. Its steady growth score of 31.60 and modest but growing number of stars suggest its utility in ensuring consistent AI behavior across different applications.
Ontos-AI/knowhere is a tool designed to extract, parse, and output structured data chunks that are ready for integration with AI agents or retrieval-augmented generation (RAG) systems. With 406 stars on GitHub, it reflects the growing demand for tools that facilitate the processing of complex datasets into easily consumable formats for AI applications.
alash3al/stash is a persistent memory layer designed to provide working context and episodes storage for AI agents using Postgres. The project includes an MCP server and can be self-hosted with minimal infrastructure, aligning well with trends towards on-premises AI solutions. With 699 stars, it indicates strong community interest in tools that enhance the persistence and state management capabilities of AI systems.
enmanuelmag/agent-harness-kit offers a scaffolding kit for running structured multi-agent workflows in various codebases without being tied to specific providers or platforms. This flexibility is highlighted by its 158 stars, suggesting developers are looking for modular tools that can be easily integrated into existing systems.
stoaaadev/stoa presents a framework for managing multi-agent swarms with autonomous agents and shared mesh networking, enabling complex interactions in decentralized environments. With 26 stars, it reflects niche interest from developers working on advanced swarm intelligence applications where such frameworks are crucial.
Lastly, TheRunicDev/MaaNTE appears to be another iteration of an automation assistant framework, possibly with a focus on game mechanics and user convenience features like auto-fishing or story dialogue skipping. Its 20.27 growth score and 431 stars indicate steady interest in automation tools that enhance gaming experiences through AI.
These projects reflect the diversity and innovation within the AI frameworks & SDKs space, from high-performance computing to multi-agent systems and beyond.