#1: tt-a1i/archify
JavaScript · 51,486 stars total, +17,190 this period
The project archify is a JavaScript-based agent skill designed to generate visually appealing and interactive diagrams for various technical and non-technical contexts. It supports the creation of architecture, workflow, sequence, data-flow, and lifecycle diagrams, all encapsulated within self-contained HTML files. These diagrams can include motion effects, making them dynamic and engaging. The tool's strength lies in its simplicity and the ability to produce high-quality, exportable visualizations directly from code, which can be particularly useful for documentation, presentations, and technical communication.
Given the significant increase in stars—17,190 gained over the current period—this project is likely trending due to its broad utility and the large community support. The high number of stars (51,486) indicates that the project has garnered substantial interest and adoption, suggesting that it fills a critical need for developers, engineers, and technical communicators seeking to enhance their diagramming capabilities. The project's focus on self-contained, interactive, and exportable diagrams aligns with current trends in digital documentation and technical communication, making it a relevant and valuable tool for the tech community.
#2: magnitudedev/magnitude
TypeScript · 3,841 stars total, +1,961 this period
Magnitude is an open-source inference engine for running open models locally. It compiles and tunes its kernels on your own hardware before a model runs, and claims up to 2x the speed of llama.cpp. It works on Apple Silicon, NVIDIA, AMD or a plain CPU, and ships as a desktop app with a bundled CLI. It connects in one click to the coding agent you already use — Pi, OpenCode, Hermes, OpenClaw, Codex, Claude Code, Oh My Pi and Cline are listed — and anything else can use its OpenAI-compatible API. ("Pi" here is the coding agent, not the Raspberry Pi.)
It gained 1,961 stars this period for a total of 3,841.
#3: THU-MAIC/OpenMAIC
TypeScript · 32,647 stars total, +9,193 this period
The project Open Multi-Agent Interactive Classroom (OpenMAIC), hosted on GitHub and written in TypeScript, aims to provide a seamless and engaging learning environment through multi-agent interaction. This initiative addresses the growing demand for interactive and immersive educational tools, allowing users to access a dynamic learning experience with minimal setup effort. The project's primary focus is on enabling educators and learners to explore complex systems and concepts through interactive simulations, fostering a more engaging and participatory learning process.
Given its impressive growth metrics, OpenMAIC is likely trending this week due to its significant increase in star counts, with 9,193 stars gained in the recent period. This rapid rise in popularity underscores the project's value and appeal, particularly among developers and educators seeking innovative tools to enhance the learning experience. The substantial number of stars also indicates a strong community backing and a growing interest in the project, suggesting that OpenMAIC is well-positioned to continue gaining traction in the educational technology space.
#4: Gitlawb/openclaude
TypeScript · 32,869 stars total, +1,944 this period
OpenClaude is an open-source coding-agent CLI that works with many model providers instead of one. It supports OpenAI-compatible APIs, Gemini, GitHub Models, Codex, Ollama and other backends, while keeping a single terminal workflow with tools, sub-agents, MCP, slash commands and streaming output. That is what its tagline, "runs anywhere. uses anything," refers to. It also bundles a VS Code extension.
It has 32,869 stars, with 1,944 gained this period. The repository has since moved to Twigpine/openclaude; the old link redirects.
#5: pollen-robotics/microduck_rl
Python · 1,853 stars total, +1,122 this period
microduck_rl contains the reinforcement learning training environments for Microduck, a small bipedal robot from Pollen Robotics (about 800 g and 25 cm tall). Walking policies are trained in simulation with MuJoCo and PPO, exported to ONNX, and then run on the real robot. The repo documents the full sim-to-real recipe, including actuator modelling, domain randomisation and the reward-design lessons that made it work. Training needs a CUDA GPU.
It gained 1,122 stars this period for a total of 1,853.