📦 openbmb/agentcpm

Build Powerful AI Agents with AgentCPM

Want to supercharge your AI agents for complex research? AgentCPM provides an end-to-end infrastructure designed to tackle long-horizon, autonomous tasks with edge-scale models. It includes AgentCPM-Explore, which leverages a 4B parameter model to handle continuous, multi-turn interactions, and AgentCPM-Report, which uses an 8B model to autonomously generate deeply insightful long-form documents from your private data. With integrated tools like the AgentDock sandbox and asynchronous reinforcement learning, you get a full suite for training and evaluation. It is a powerful, open-source solution for running sophisticated, private-first research agents directly on your own hardware.

🆔 @hackernewsgithubprojects

📦 shanxierdan/dibs

Supercharge Your Solvers with Diffusion-Informed Branch Selection

Struggling with slow symbolic solvers on tough problems? Meet DiBS, a clever new approach that blends the accuracy of traditional solvers with the intelligence of machine learning. Instead of relying on rigid search patterns, DiBS uses a learned diffusion model to intelligently reorder branches during the search process. This boosts performance on complex tasks like Sudoku or 3-SAT by prioritizing better candidates while keeping your solver complete and reliable. It is the perfect bridge between raw computational speed and smart, data-driven reasoning. Give your solvers a massive upgrade and start tackling those long-tail search problems with much greater efficiency today.

🆔 @hackernewsgithubprojects

📦 tianyi-lab/polar

Optimize LLM Inference by Skipping or Looping Layers

Did you know your favorite AI model might be doing unnecessary work? Large language models usually run every single layer, but the PoLar framework proves we can achieve better results by dynamically skipping or looping through specific layers for each input. This clever approach treats layer execution like a customizable program, allowing for shorter, more efficient inference paths that maintain or even improve accuracy. By using a lightweight prediction network to decide the best path, you can unlock deeper reasoning capabilities hidden within standard models. It is a smarter way to run AI that makes your models faster and more effective.

🆔 @hackernewsgithubprojects

📦 marswang42/orbitos

OrbitOS: Your AI-Orchestrated Personal Productivity System

Are you tired of manually organizing your tasks and notes? Meet OrbitOS, an Obsidian-based framework that turns your personal productivity into an intelligently orchestrated system. By leveraging tools like Claude or Gemini, it acts as a dedicated AI partner that manages your knowledge, plans your daily routine, and helps you execute projects. Simply capture ideas into your inbox, and use commands like kickoff or research to structure them into actionable plans. With features for automated knowledge parsing, project tracking, and deep research, it keeps your work moving forward effortlessly. Reclaim your focus and let your AI assistant organize your life.

🆔 @hackernewsgithubprojects

📦 cytostack/openwolf

Slash Your Claude Code Tokens with OpenWolf

Are you tired of Claude Code burning through your token budget by reading the same files over and over? Meet OpenWolf, the open-source middleware that gives your AI agent a second brain. It works by injecting invisible hooks into your workflow, creating a project map, a learning memory, and a token ledger. By letting Claude know what is in your files before it opens them, OpenWolf significantly reduces unnecessary reads and saves massive amounts of tokens. It even tracks your project anatomy and bug history to keep your agent sharp. Take control of your AI costs and optimize your development flow today.

🆔 @hackernewsgithubprojects

📦 dragonflydb/dragonfly

Is This the Ultimate Redis Alternative?

Are you struggling with the limitations of traditional in-memory data stores? Meet Dragonfly, a modern, high-performance replacement for Redis and Memcached designed for today’s demanding application workloads. By utilizing a unique multi-threaded, shared-nothing architecture, Dragonfly delivers up to twenty-five times the throughput of older systems while maintaining significantly lower tail latency. It is fully compatible with existing Redis and Memcached APIs, meaning you can switch over without changing your code. Whether you need massive scalability or better memory efficiency during heavy tasks like data snapshots, Dragonfly offers a powerful, seamless upgrade for your infrastructure.

🆔 @hackernewsgithubprojects

📦 gszhangwei/open-spdd

Stop AI from Guessing: Use OpenSPDD to Control Your Code

Are you tired of AI coding assistants hallucinating or writing code that ignores your requirements? Meet OpenSPDD, a powerful CLI tool that transforms your prompts into executable design contracts. Instead of leaving AI to improvise, it uses a seven-dimensional REASONS Canvas framework to define everything from domain entities to strict coding guardrails. This tool supports popular environments like Cursor and GitHub Copilot, allowing you to generate code that strictly follows your design and even syncs changes back automatically. Take command of your development workflow and ensure your design and implementation finally stay perfectly in sync.

🆔 @hackernewsgithubprojects

📦 owenselles/cloudnow

Native GeForce NOW on Apple TV!

Want to stream your entire PC game library on your Apple TV without relying on a browser or clunky workarounds? CloudNow is an impressive open-source project that delivers a native GeForce NOW experience directly to your television. By reverse-engineering the official streaming protocol, this app lets you browse your games, manage your library, and jump into high-quality sessions with full controller support and even microphone integration. It handles everything from server region selection to real-time stats overlays, providing a polished way to game on your big screen. Check out the project to start building your own native streaming hub today.

📰 https://news.ycombinator.com/item?id=48488423

🆔 @hackernewsgithubprojects

📦 m-schuetz/curast

Rendering Billions of Triangles with CUDA

Ever wonder how to render billions of triangles in real time? Meet a breakthrough software project that challenges standard hardware pipelines by using custom compute shaders. Instead of relying on traditional methods, it employs a smart three-stage rasterization pipeline tailored for massive datasets like dense photogrammetry meshes. By processing small triangles with single threads and scaling up compute power for larger ones, this approach delivers incredible performance gains over conventional rendering. While still evolving, it proves that shifting heavy geometry processing to highly efficient CUDA kernels can outperform traditional hardware, setting a new bar for future real-time graphics.

🆔 @hackernewsgithubprojects

📦 huggingface/lelab

Master Your Robotics Projects with LeLab

Struggling with complex command-line setups for your robotic arms? Say hello to LeLab, the official graphical interface for the LeRobot framework. This powerful web application replaces tricky keyboard prompts and terminal gymnastics with an intuitive browser experience designed for efficiency. Whether you need to calibrate hardware, teleoperate your robot, record datasets, or launch training jobs, LeLab puts every step into a clean, unified dashboard. It is a fantastic way for newcomers and experts alike to simplify their robotics workflow and start training policies in minutes. Dive into the interface and take full control of your automation projects today.

🆔 @hackernewsgithubprojects

📦 jasonkneen/tiny-world-builder

Build Your Own Tiny Worlds with Code

Have you ever wanted to build your own interactive 3D world right in your browser? Tiny World Builder is a fascinating project that lets you design isometric landscapes using simple tools like terrain painting and object placement. It uses a custom engine to handle everything from procedural road networks and animated traffic to house clusters that automatically connect as you build. Whether you are creating a small village or testing large-scale vehicle simulations, this project shows how to turn basic primitives into a dynamic, living scene. Dive into the code to start building your own miniature world today.

🆔 @hackernewsgithubprojects

📦 thudm/slime

Slime: The Secret Engine Behind Large-Scale RL Post-Training

Ever wonder how top-tier AI labs train their most advanced models with reinforcement learning? Meet slime, a specialized framework built for high-performance RL scaling. By tightly connecting the Megatron training stack with the SGLang inference engine, slime handles the heavy lifting of massive model updates and high-throughput data generation in one unified pipeline. It eliminates the need for complex, disconnected systems, letting you run efficient training and flexible data workflows seamlessly. From agentic tool use to large-scale reasoning, slime provides the robust infrastructure needed to push model performance further. Master your RL scaling today and level up your production AI.

🆔 @hackernewsgithubprojects

📦 daronyondem/claude-architect-exam-guide

Ace Your Claude Architect Exam!

Are you gearing up for the Claude Certified Architect Foundations exam? If you are feeling overwhelmed, this community study guide is exactly what you need to streamline your preparation. It covers critical concepts like API fundamentals, agentic patterns, and the Model Context Protocol, while helping you master essential design trade-offs for production-ready applications. This project acts as your comprehensive roadmap, providing structured insights, common pitfalls, and clear strategies to build your expertise. Instead of just memorizing facts, you will gain the deep understanding required to excel. Grab this resource today and start building your confidence for exam day.

🆔 @hackernewsgithubprojects

📦 openocta/openocta

OpenOcta: Enterprise-Grade AI Agent Runtime

Struggling to integrate AI into your enterprise workflow? Meet OpenOcta, an open-source, enterprise-grade AI agent designed for team automation. Built as a single Go binary with an embedded frontend, it acts as a unified control plane that simplifies how you connect business systems, APIs, and custom tools. Whether you need workflow automation, task orchestration, or a robust gateway for your agents, OpenOcta is built to deploy anywhere. With features like built-in Chromium support for browser tasks and seamless integration with existing toolchains, it makes managing complex enterprise workflows feel effortless. Level up your team’s productivity and take full control of your AI infrastructure today.

🆔 @hackernewsgithubprojects

📦 jimliu/baoyu-design

Design Like a Pro Inside Your Editor with Baoyu-Design

Imagine bringing professional design power directly into your code editor. Baoyu-design allows you to run high-end design engines locally as an agent skill within environments like Cursor or Claude Code. Instead of relying on web tools, you can now generate polished UI mockups, interactive prototypes, and even slide decks as self-contained HTML files right in your repository. Whether you are building mobile apps or complex dashboards, this project lets you iterate visually by pointing at elements in your preview to make live edits. It keeps your design assets organized, version-controlled, and completely under your control. Take your creative workflow to the next level today.

🆔 @hackernewsgithubprojects

📦 tencent-hunyuan/hunyuan-a13b

Meet Hunyuan-A13B: A High-Efficiency Powerhouse LLM

Are you looking for top-tier AI performance without needing a supercomputer? Meet Hunyuan-A13B, an innovative open-source large language model that solves the trade-off between power and resource consumption. Built on a fine-grained Mixture-of-Experts architecture, it features eighty billion total parameters but only thirteen billion active ones. This design delivers impressive reasoning, coding, and agent capabilities while remaining highly efficient for various environments. It even supports an ultra-long context window of 256K and offers flexible thinking modes for complex tasks. Whether for research or cost-effective AI development, Hunyuan-A13B is a versatile foundation designed to make advanced artificial intelligence more accessible.

🆔 @hackernewsgithubprojects

📦 disler/claude-code-hooks-mastery

Mastering Claude Code Hooks: Ultimate Guide

Ever wanted to take full control over your AI coding assistant? This repository is a masterclass in using Claude Code hooks to customize your workflow. It shows you how to implement deterministic control, like blocking dangerous commands, validating prompts, or even triggering voice notifications. By using Python scripts managed with UV, you can securely intercept tool usage and monitor your AI agent lifecycle in real time. It even features advanced patterns like team-based validation and custom slash commands to streamline your development. Dive in to transform your coding environment into a powerful, automated, and truly personalized AI workspace.

🆔 @hackernewsgithubprojects

📦 wquguru/harness-books

Master AI Agent Design with Harness Engineering

Ever wonder how to keep AI coding agents from going off the rails? Harness engineering is the secret to building stable, accountable systems. This repository offers two comprehensive books that dive deep into the design philosophies behind tools like Claude Code and Codex. Instead of focusing on basic prompts, these books explore essential structural components like query loops, permission systems, and multi-agent verification that turn unstable models into reliable, professional workflows. Whether you are building your own harness or just want to understand the architecture behind top-tier coding agents, these guides provide the frameworks you need to succeed.

🆔 @hackernewsgithubprojects

📦 viperrcrypto/siftly

Turn Your Messy Twitter Bookmarks Into An AI-Powered Library

Are your Twitter bookmarks a disorganized mess you never look at again? Siftly is a self-hosted tool that transforms your bookmarks into a searchable, visual knowledge base directly on your machine. It runs a powerful four-stage AI pipeline, performing entity extraction, vision analysis on media, and semantic tagging to categorize everything for you. You can explore your collection through an interactive mindmap or use natural language to find exactly what you need. It even supports Claude CLI for automatic authentication, keeping your data private and local. Take control of your saved content and make it useful again today.

🆔 @hackernewsgithubprojects

📦 shpigford/chops

Stop Managing Your AI Agent Skills Manually!

Are you tired of digging through hidden system files just to manage your AI agent skills? Meet Chops, a native macOS application designed to finally bring order to your coding workflow. Chops acts as a central hub where you can discover, edit, and organize skills and agents across tools like Claude Code, Cursor, and Windsurf. With features like real-time file watching, full-text search, and a built-in markdown editor, it replaces tedious manual configuration with a smooth, native interface. Stop manually editing dotfiles and start managing your AI agents effectively today. It is the perfect way to streamline your development process.

🆔 @hackernewsgithubprojects