Today's AI Frameworks & SDKs: Fastest-Growing Projects — May 21, 2026
Today's the AI Frameworks & SDKs space, we see a diverse range of projects gaining traction across various domains such as finance, gaming, and security. One notable trend is the rise of specialized frameworks designed to enhance efficiency and performance for large language models (LLMs) and multi-agent systems. Leading Today's list is "TokenSpeed," which stands out for its unique approach to speeding up LLM inference.
lightseekorg/tokenspeed
TokenSpeed is a speed-of-light LLM inference engine, aiming to drastically reduce the time required for processing large language models. With over 1,069 stars and a robust growth score of 65.33, it demonstrates significant interest from developers looking to optimize their AI applications' performance.
simonlin1212/TradingAgents-astock
This project offers a multi-agent investment research framework tailored for the A-share market in China, integrating various data sources and enabling complex decision-making processes among seven AI analysts. Its growth score of 57.12 and 437 stars reflect its growing popularity among financial technology enthusiasts seeking advanced trading strategies.
1bananachicken/MaaNTE
MaaNTE is described as a powerful assistant for the Neverness to Everness game, featuring automation capabilities such as auto-fishing, selling fish, buying bait, making coffee, and more. With 1,428 stars and a growth score of 48.18, it showcases strong community engagement and utility among gamers looking to enhance their in-game experience.
jhaizhou-ops/pinrule
PinRule is a universal AI behavior rule framework that allows users to "pin" specific rules for AI systems to follow during long tasks, ensuring consistency without the need for additional language models or network dependencies. Its steady growth score of 39.50 and 29 stars indicate its relevance in maintaining stable AI behaviors across various applications.
noonghunna/club-3090
Club-3090 provides community-driven recipes for running large language models on NVIDIA RTX 3090 GPUs, offering multi-engine support and model agnosticism. With a growth score of 38.41 and 1,017 stars, it highlights the demand for optimized hardware configurations to enhance LLM performance.
TheRunicDev/MaaNTE
Another MaaNTE entry focuses on automation within the Neverness to Everness game environment, offering features like auto-fishing, bait buying, and daily reward claiming. Its 600 stars and growth score of 34.17 reflect its utility for players seeking automated gameplay enhancements.
stoaaadev/stoa
Stoa is a multi-agent swarm framework designed to manage up to seven autonomous agents with 30 skills operating on a single shared mesh network, requiring no additional infrastructure setup. With 23 stars and a growth score of 32.83, it suggests interest from developers looking for scalable and efficient agent-based systems.
Ontos-AI/knowhere
Knowhere is an AI-driven tool that extracts and structures data into chunks ready for further processing by AI agents or retrieval-augmented generation (RAG) systems. Its growth score of 24.83 and 338 stars point to its growing utility in enhancing the interoperability between different AI components.
alash3al/stash
Stash offers a persistent memory layer designed for AI agents, storing episodes, facts, and working context within PostgreSQL databases. With a growth score of 24.54 and 698 stars, it underscores the demand for robust data storage solutions in AI applications.
OmYarewar/PHANTOM
Phantom is an AI-powered pentesting command center that supports autonomous security testing with real-time streaming capabilities and a self-improving AI system. Its growth score of 18.95 and 39 stars indicate its relevance to cybersecurity professionals seeking advanced, automated penetration testing tools.
These projects highlight the growing diversity and specialization within the AI frameworks and SDKs space, catering to varied needs from gaming automation to financial analysis and security testing.
lightseekorg/tokenspeed
TokenSpeed is a speed-of-light LLM inference engine, aiming to drastically reduce the time required for processing large language models. With over 1,069 stars and a robust growth score of 65.33, it demonstrates significant interest from developers looking to optimize their AI applications' performance.
simonlin1212/TradingAgents-astock
This project offers a multi-agent investment research framework tailored for the A-share market in China, integrating various data sources and enabling complex decision-making processes among seven AI analysts. Its growth score of 57.12 and 437 stars reflect its growing popularity among financial technology enthusiasts seeking advanced trading strategies.
1bananachicken/MaaNTE
MaaNTE is described as a powerful assistant for the Neverness to Everness game, featuring automation capabilities such as auto-fishing, selling fish, buying bait, making coffee, and more. With 1,428 stars and a growth score of 48.18, it showcases strong community engagement and utility among gamers looking to enhance their in-game experience.
jhaizhou-ops/pinrule
PinRule is a universal AI behavior rule framework that allows users to "pin" specific rules for AI systems to follow during long tasks, ensuring consistency without the need for additional language models or network dependencies. Its steady growth score of 39.50 and 29 stars indicate its relevance in maintaining stable AI behaviors across various applications.
noonghunna/club-3090
Club-3090 provides community-driven recipes for running large language models on NVIDIA RTX 3090 GPUs, offering multi-engine support and model agnosticism. With a growth score of 38.41 and 1,017 stars, it highlights the demand for optimized hardware configurations to enhance LLM performance.
TheRunicDev/MaaNTE
Another MaaNTE entry focuses on automation within the Neverness to Everness game environment, offering features like auto-fishing, bait buying, and daily reward claiming. Its 600 stars and growth score of 34.17 reflect its utility for players seeking automated gameplay enhancements.
stoaaadev/stoa
Stoa is a multi-agent swarm framework designed to manage up to seven autonomous agents with 30 skills operating on a single shared mesh network, requiring no additional infrastructure setup. With 23 stars and a growth score of 32.83, it suggests interest from developers looking for scalable and efficient agent-based systems.
Ontos-AI/knowhere
Knowhere is an AI-driven tool that extracts and structures data into chunks ready for further processing by AI agents or retrieval-augmented generation (RAG) systems. Its growth score of 24.83 and 338 stars point to its growing utility in enhancing the interoperability between different AI components.
alash3al/stash
Stash offers a persistent memory layer designed for AI agents, storing episodes, facts, and working context within PostgreSQL databases. With a growth score of 24.54 and 698 stars, it underscores the demand for robust data storage solutions in AI applications.
OmYarewar/PHANTOM
Phantom is an AI-powered pentesting command center that supports autonomous security testing with real-time streaming capabilities and a self-improving AI system. Its growth score of 18.95 and 39 stars indicate its relevance to cybersecurity professionals seeking advanced, automated penetration testing tools.
These projects highlight the growing diversity and specialization within the AI frameworks and SDKs space, catering to varied needs from gaming automation to financial analysis and security testing.