Issue #84 · 2025-10-21
sing-box is a cutting-edge proxy platform written in Go that provides universal proxy support. With its open-source nature and robust feature set, it competes with proprietary and expensive proxy solutions. If you're looking for a powerful, flexible, and community-driven proxy solution, sing-box is worth exploring.
github_trending · 2 min read
Mindustry is a unique, cross-platform automation-focused tower defense and real-time strategy (RTS) game developed in Java. Its open-source nature allows for deep customization, modding, and potential integration into educational or simulation projects. If you're a gamer or a developer interested in game modding, Mindustry could be a great fit.
hackernews · 3 min read
LeRobot is a groundbreaking library for AI-powered robotics, offering accessible end-to-end learning solutions. It provides pretrained models, datasets, and tools focused on real-world applications. For developers working in robotics or AI, LeRobot is a must-check.
reddit · 4 min read
Atlas is an open-source modification to Windows that debloats the OS and optimizes performance, privacy, and security without breaking essential features. It leverages community-driven transparency via GitHub. If you're a Windows user looking to enhance your system, Atlas OS might be the answer.
hackernews · 3 min read
GPU-poor introduces a Python library that applies INT8 quantization to large language models, compressing them by 74% with minimal loss in quality. This enables running models on memory-constrained devices. If you're working with LLMs and facing hardware limitations, this could be a game-changer.
AI/ML
Python · 1200 stars
ToolCritic introduces a diagnostic framework for detecting and correcting tool-use errors in dialogue systems. It identifies eight specific error types, provides targeted feedback, and improves tool-calling accuracy. If you're developing or maintaining dialogue systems, this could help enhance their reliability.
NLP/AI
Python · 500 stars
This paper introduces STARK, a multi-agent LLM framework for GPU kernel optimization. It addresses key challenges in exploration, agent design, and the planning-implementation gap. If you're into GPU optimization or multi-agent systems, this research is a must-read.
AI/ML · Yuxi Wang et al. · 5 min read
This research introduces VAGEN, a framework that enables multi-turn VLM agents to build accurate internal world models through explicit visual state and transition reasoning. By structuring the agent's reasoning process, it improves performance in multi-turn visual interactions. If you're working with VLMs or multi-turn agents, this is relevant.
AI/ML · Yiwen Chen et al. · 5 min read