Today's AI Frameworks & SDKs: Fastest-Growing Projects — May 20, 2026
Today's the AI Frameworks & SDKs space, there's a notable surge of interest in projects that focus on optimizing large language models (LLMs) and enhancing their performance through multi-agent collaboration or rule-based behavior constraints. One standout project is lightseekorg/tokenspeed, which offers an impressive growth score and has gained significant traction among developers looking for efficient LLM inference solutions.
TokenSpeed by lightseekorg is a speed-of-light LLM inference engine designed to significantly enhance the performance of language models. With its high Growth Score of 69.82 and over 1,000 stars, it clearly resonates with users seeking faster and more efficient ways to run large-scale AI applications.
Simonlin1212's TradingAgents-astock is an A-share multi-agent investment research framework that integrates various data sources relevant to the Chinese stock market for AI-driven analysis. The project boasts a Growth Score of 61.29, likely due to its unique approach and relevance in the Asian financial technology sector.
MaaNTE by 1bananachicken is an automation assistant powered by MAAFramework, designed for specific game mechanics such as fishing and trading within virtual environments. With over 1,400 stars, this project's Growth Score of 49.96 highlights its popularity among users interested in automated gameplay solutions.
Jhaizhou-ops' pinrule is a universal AI behavior rule framework aimed at preventing AI systems from deviating from predefined tasks by enforcing specific rules without requiring network connectivity or additional dependencies. The project’s steady growth, as indicated by its Growth Score of 45.14 and consistent commits, suggests it meets a critical need in the field of AI governance.
Noonghunna's club-3090 provides community recipes for serving LLMs on RTX 3090 GPUs, supporting multiple engines and models without being model-specific. Its Growth Score of 40.09 reflects its utility in optimizing hardware resources for large-scale language model deployments.
TheRunicDev’s MaaNTE is another instance of an automation assistant, this time focusing on the Neverness to Everness game environment with features like auto-fishing and revenue extraction. Despite a lower Growth Score of 36.75 compared to its peers, it has attracted over 500 stars, indicating strong niche interest among gaming enthusiasts.
Harmonist-orchestral by 2508965-ship-it is a multi-agent orchestration engine built for deploying AI swarms using Claude Code. The project's Growth Score of 35.67 suggests it’s gaining traction as developers look to manage complex AI systems across various tasks and environments.
Alash3al's stash introduces a persistent memory layer for AI agents, storing episodes, facts, and working context in Postgres. With over 600 stars and a Growth Score of 25.46, the project demonstrates its value as an essential tool for developers aiming to enhance AI agent capabilities with long-term memory.
Ontos-AI's knowhere is designed for extracting, parsing, and outputting structured data chunks suitable for use by AI agents in retrieval-augmented generation (RAG) tasks. Its Growth Score of 25.30 reflects its utility in preparing datasets that can be leveraged across different AI frameworks.
ZhiYi-R's moon-bridge is a forwarding layer designed to convert messages from Anthropic’s API format into OpenAI-compatible responses, facilitating easier integration between different provider APIs. With a Growth Score of 18.82 and over 200 stars, it addresses a specific need in the rapidly evolving AI ecosystem where interoperability is crucial.
These projects collectively showcase the diversity and innovation within the AI Frameworks & SDKs space, with developers focusing on performance optimization, multi-agent collaboration, rule-based governance, hardware optimization, persistent memory management, and API compatibility.
TokenSpeed by lightseekorg is a speed-of-light LLM inference engine designed to significantly enhance the performance of language models. With its high Growth Score of 69.82 and over 1,000 stars, it clearly resonates with users seeking faster and more efficient ways to run large-scale AI applications.
Simonlin1212's TradingAgents-astock is an A-share multi-agent investment research framework that integrates various data sources relevant to the Chinese stock market for AI-driven analysis. The project boasts a Growth Score of 61.29, likely due to its unique approach and relevance in the Asian financial technology sector.
MaaNTE by 1bananachicken is an automation assistant powered by MAAFramework, designed for specific game mechanics such as fishing and trading within virtual environments. With over 1,400 stars, this project's Growth Score of 49.96 highlights its popularity among users interested in automated gameplay solutions.
Jhaizhou-ops' pinrule is a universal AI behavior rule framework aimed at preventing AI systems from deviating from predefined tasks by enforcing specific rules without requiring network connectivity or additional dependencies. The project’s steady growth, as indicated by its Growth Score of 45.14 and consistent commits, suggests it meets a critical need in the field of AI governance.
Noonghunna's club-3090 provides community recipes for serving LLMs on RTX 3090 GPUs, supporting multiple engines and models without being model-specific. Its Growth Score of 40.09 reflects its utility in optimizing hardware resources for large-scale language model deployments.
TheRunicDev’s MaaNTE is another instance of an automation assistant, this time focusing on the Neverness to Everness game environment with features like auto-fishing and revenue extraction. Despite a lower Growth Score of 36.75 compared to its peers, it has attracted over 500 stars, indicating strong niche interest among gaming enthusiasts.
Harmonist-orchestral by 2508965-ship-it is a multi-agent orchestration engine built for deploying AI swarms using Claude Code. The project's Growth Score of 35.67 suggests it’s gaining traction as developers look to manage complex AI systems across various tasks and environments.
Alash3al's stash introduces a persistent memory layer for AI agents, storing episodes, facts, and working context in Postgres. With over 600 stars and a Growth Score of 25.46, the project demonstrates its value as an essential tool for developers aiming to enhance AI agent capabilities with long-term memory.
Ontos-AI's knowhere is designed for extracting, parsing, and outputting structured data chunks suitable for use by AI agents in retrieval-augmented generation (RAG) tasks. Its Growth Score of 25.30 reflects its utility in preparing datasets that can be leveraged across different AI frameworks.
ZhiYi-R's moon-bridge is a forwarding layer designed to convert messages from Anthropic’s API format into OpenAI-compatible responses, facilitating easier integration between different provider APIs. With a Growth Score of 18.82 and over 200 stars, it addresses a specific need in the rapidly evolving AI ecosystem where interoperability is crucial.
These projects collectively showcase the diversity and innovation within the AI Frameworks & SDKs space, with developers focusing on performance optimization, multi-agent collaboration, rule-based governance, hardware optimization, persistent memory management, and API compatibility.