📦 graykode/abtop

Monitor Your AI Coding Agents with Abtop

Are you running multiple AI coding agents at once? Managing their tokens, rate limits, and background processes can quickly become a headache. Meet abtop, the terminal-based monitor designed specifically for tools like Claude Code and Codex. It gives you a real-time view of every active session, including context window usage, active ports, and even orphan process detection, all in one clean dashboard. It works across macOS, Linux, and Windows with no sensitive data access required. Whether you are juggling multiple projects or just want to stay within your limits, abtop helps you stay in complete control.

🆔 @hackernewsgithubprojects

📦 eunomia-bpf/agentsight

Stop Guessing What Your AI Agents Are Doing

Ever wonder what AI agents are actually doing to your computer behind the scenes? Meet AgentSight, a powerful tool that monitors AI agents at the system level without needing any code changes or tricky integrations. By using eBPF, it tracks everything from subprocess executions and file operations to network activity and even decrypted LLM prompts, all without relying on SDKs or proxies. Whether you are using open or closed-source CLI agents, AgentSight provides a clear, real-time view of their behavior and system impact. Install it today to finally bridge the gap between AI decisions and your machine’s actual performance.

📰 https://news.ycombinator.com/item?id=48386737

🆔 @hackernewsgithubprojects

📦 openai/imagegencam

Build Your Own AI-Powered Camera

Ever wanted to build a digital camera that transforms your world into art? Meet ImageGenCam, an open-source project that lets you create a custom point-and-shoot camera using a Raspberry Pi and a 3D-printed case. By integrating generative AI, this device takes your photos and instantly reimagines them through unique prompts, like turning your subjects into cheese or stylized anime characters. It is a fantastic weekend project that blends hardware engineering with creative coding. Grab your parts, use the provided assistant to guide your assembly, and start building your own programmable camera to capture the world in your own style.

📰 https://news.ycombinator.com/item?id=48372591

🆔 @hackernewsgithubprojects

📦 duanebester/gooey

Build High-Performance GUIs with Gooey

Stop struggling with sluggish user interfaces and meet Gooey, a powerful hybrid framework built for the Zig programming language. This project enables you to create lightning-fast, GPU-rendered applications that run seamlessly across macOS, Linux, and the web. By utilizing technologies like Metal, Vulkan, and WebGPU, Gooey provides developers with a robust, component-based system that blends immediate and retained mode rendering for optimal performance. Whether you need complex animations, built-in themes, or native file dialogs, this framework simplifies the process while keeping your code clean and testable. Dive into Gooey today to start building fluid, professional-grade interfaces with ease.

📰 https://news.ycombinator.com/item?id=48386725

🆔 @hackernewsgithubprojects

📦 kuba-guzik/caveman-micro

Slash AI Token Costs with This 6-Line Prompt

Are you tired of wasting expensive tokens on overly wordy AI responses? Meet Caveman Micro, an incredibly efficient six-line prompt that forces any LLM to get straight to the point without sacrificing quality. While other prompts use hundreds of tokens to explain conciseness, this tiny instruction set cuts the fluff, drops articles, and removes unnecessary pleasantries to save you significant processing costs. Tested on Claude Sonnet and Opus, it maintains perfect accuracy while maximizing efficiency. Give your AI permission to be brief today, save on those long-term API calls, and start optimizing your workflow with this simple, powerful shortcut.

📰 https://news.ycombinator.com/item?id=47691809

🆔 @hackernewsgithubprojects

📦 bergside/design-md-figma

Turn Figma Designs into AI-Ready Blueprints

Tired of design inconsistencies when coding? This Figma plugin automatically extracts your local styles and variables to create a structured design and skill markdown file. By generating these blueprints, you provide AI coding assistants with clear, actionable context on your colors, typography, and component rules. It eliminates guesswork for your agents, ensuring they follow your design system perfectly every time. Whether you are building complex interfaces or simple sites, this tool bridges the gap between your design files and AI development workflows. Try it out to finally sync your design and code effortlessly starting today.

📰 https://news.ycombinator.com/item?id=47924396

🆔 @hackernewsgithubprojects

📦 hanxiao/dataroom

Automate Your Research with Dataroom

Stop wasting expensive frontier model tokens on repetitive research tasks. Dataroom is a powerful local research harness that uses a self-hosted Qwen model to autonomously search, read, and write comprehensive data. It executes a disciplined loop of gathering and organizing information until it produces a structured zip file containing fully-cited summaries and raw data. This output is perfectly designed as context for a second, more advanced AI agent to tackle complex projects. By handling the heavy lifting locally on your own GPU, Dataroom ensures your research is thorough and cost-effective. Streamline your AI pipeline and start researching smarter today.

📰 https://news.ycombinator.com/item?id=48362986

🆔 @hackernewsgithubprojects

📦 bernardohcrocha/persistia-for-claude-code

Give Claude Code a Brain with Persistia

Are you tired of constantly re-explaining your project to Claude Code every single time you start a new session? Meet Persistia, the tool that gives your coding agent a persistent, self-updating memory. By scanning your files and tracking changes through git, it builds a private knowledge base that stays perfectly in sync with your work. From storing your permanent skills to scheduling automated tasks, it transforms your agent into a true operational co-pilot that remembers everything. Stop wasting time on context setup and start getting more done with an assistant that finally knows your project as well as you do.

📰 https://news.ycombinator.com/item?id=48358513

🆔 @hackernewsgithubprojects

📦 autolabhq/autolab

Is Your AI Agent Actually Smart? Meet AutoLab

Are your AI agents truly capable of solving complex research problems or just guessing? AutoLab changes the game by providing a rigorous benchmark for evaluating agents on twenty-three open-ended challenges. Instead of simple one-shot answers, it forces agents to diagnose bottlenecks, run iterative experiments, and optimize performance across system-level tasks and model development projects. From accelerating C code to fine-tuning large language models, AutoLab tests the practical skills that actually matter in modern research. If you want to push the boundaries of agent capability and see if your models can handle real-world engineering constraints, this is your ultimate testing ground.

🆔 @hackernewsgithubprojects

📦 xbpeng/mimickit

Master Robot Motion with MimicKit

Have you ever wondered how robots learn to move like humans or animals? Meet MimicKit, a lightweight framework designed to help you train motion controllers with ease. It solves the complexity of motion imitation by offering a clean suite of methods like DeepMimic and Adversarial Motion Priors. Whether you are using Isaac Gym or other simulation engines, this codebase keeps dependencies minimal while supporting distributed training on multiple GPUs. It is a fantastic tool for researchers or enthusiasts looking to bridge the gap between simulation and realistic movement. Start training your own motion models today and bring your virtual agents to life.

🆔 @hackernewsgithubprojects

📦 ideogram-oss/ideogram4

Meet Ideogram 4: The New King of Open Image Models

Imagine having a professional design assistant that masters typography and photorealism instantly. Ideogram 4 has arrived as a powerful new open-weight text-to-image model built entirely from scratch. Unlike standard tools, it features a unique structured JSON prompting interface, offering precise control over color palettes and bounding-box layouts. With native two-thousand resolution support and best-in-class multilingual text rendering, it is currently leading benchmarks for open-weight models. Whether you are creating complex posters or realistic photos, this foundation model is designed to push the boundaries of visual intelligence. Discover the future of AI design by experimenting with this release today.

📰 https://news.ycombinator.com/item?id=48385829

🆔 @hackernewsgithubprojects

📦 adscanpro/adscan

Master Active Directory Pentesting on Linux with ADscan

Stop juggling multiple tools for your Active Directory pentesting because there is a better way. ADscan is a powerful Linux command line interface designed to streamline your entire security workflow into one unified experience. It replaces complex toolchains by covering over forty distinct attack techniques, including Kerberoasting, AS-REP roasting, and automated ADCS exploitation, without ever needing a Windows machine. From unauthenticated discovery to mapping complex privilege escalation paths, it handles your entire assessment within a single terminal session. Level up your security testing today by simplifying your toolkit and accelerating your penetration testing results.

🆔 @hackernewsgithubprojects

📦 nvidia/dreamdojo

DreamDojo: Teaching Robots Through Human Video

Ever wondered how robots can learn to understand our complex world? Meet DreamDojo, a groundbreaking foundation world model that learns directly from massive amounts of human video. By leveraging forty-four thousand hours of egocentric footage, it tackles the challenge of generalization across diverse objects and environments. This repository provides the pretraining and post-training code, along with powerful checkpoints to get you started. Thanks to a specialized distillation pipeline, the model achieves stable, long-horizon generation at ten frames per second. It is a massive step forward for physical AI, proving that observing human actions is key to mastering robotic interaction.

🆔 @hackernewsgithubprojects

📦 evo-hq/evo

Evo: Automate Your Code Optimization with AI Research Loops

Want to supercharge your codebase without lifting a finger? Evo is a powerful tool that turns your projects into an autonomous research loop, helping you optimize code through smart experimentation. Instead of manual testing, Evo uses parallel subagents to discover metrics, create benchmarks, and run iterative tree searches to improve performance. It intelligently handles failures, uses gates to maintain code quality, and provides a dashboard to monitor progress across your experiments. Whether you need to speed up a parser or improve efficiency, Evo automates the trial and error process for you. Start optimizing your entire workflow today.

🆔 @hackernewsgithubprojects

📦 swe-agent/mini-swe-agent

Build Powerful AI Agents in 100 Lines of Code

Ever wondered how to build a high-performance AI coding agent without the massive complexity? Meet the mini-swe-agent, a radically simple tool that packs incredible power into just a hundred lines of code. Unlike bloated alternatives, this agent ditches complex scaffolds and custom tools, relying entirely on bash to interact with your environment. It handles tasks independently using subprocess calls, making it surprisingly stable and easy to debug. It even scores over seventy-four percent on benchmark tests. Whether you are a researcher or a developer, this minimal design puts the model in control, offering a truly hackable path to better coding automation.

🆔 @hackernewsgithubprojects

📦 snailsploit/claude-red

Turn Claude Into a Red Team Expert

Want to supercharge your AI for cybersecurity tasks? Check out Claude-Red, a curated library of offensive security skills designed for the Claude system. This project provides structured files that prime Claude with expert-level methodologies across diverse attack surfaces like web exploits, EDR evasion, and exploit development. By dropping these skills into your environment, Claude begins to act like a specialized red team operator, guiding you through complex security challenges with precision. It is a powerful resource for bug bounty hunters and researchers looking to explore attack paths methodically. Give your AI the ultimate security upgrade today.

🆔 @hackernewsgithubprojects

📦 awizemann/scarf

Take Control of Your AI Agent with Scarf

Want total visibility into your AI agent? Meet Scarf, the native macOS and iOS companion app built specifically for the Hermes AI agent. It gives you a clean interface to monitor exactly what your agent is doing, when it acts, and what it creates. Whether you are managing multi-server setups, tweaking project-scoped chats, or auditing your supply chain, Scarf puts the power in your hands. With features like Kanban board management, secret storage, and secure remote access over SSH, it is the ultimate toolkit for power users. Elevate your AI workflow today and see what your agent is really building!

🆔 @hackernewsgithubprojects

📦 capgym/cap-x

Mastering Robot Control with CaP-X

Ever wondered how we can teach robots to think and act using code? Meet CaP-X, a powerful open-access framework designed to benchmark and improve coding agents for robot manipulation. It provides interactive environments where agents control robots by generating Python code, alongside a systematic benchmark to evaluate performance across various tasks. The project even includes a training-free agentic framework and reinforcement learning capabilities for post-training language models to bridge the gap between simulation and the real world. Whether you are experimenting with perception or motion planning, CaP-X gives you the tools to advance intelligent robotics today.

🆔 @hackernewsgithubprojects

📦 adrianhajdin/ai-resume-analyzer

Build an AI Resume Analyzer with React and Puter.js

Ever wonder how to build an intelligent app without a complex backend? This project shows you how to create an AI-powered resume analyzer using React and Puter.js. It simplifies the hiring process by letting users securely upload resumes and get instant AI-driven evaluations and ATS scores tailored to specific job listings. By leveraging Puter.js, you can handle authentication and storage directly in the browser, keeping your development process smooth and efficient. It is a fantastic way to master modern web tools and build a professional, reusable interface. Dive into the code to level up your development skills today.

🆔 @hackernewsgithubprojects

📦 artificialanalysis/stirrup

Build Better AI Agents with Stirrup

Are you tired of rigid AI frameworks that restrict how your models solve complex tasks? Meet Stirrup, a lightweight foundation designed to let your AI agents think for themselves. Unlike restrictive alternatives, it gives models the freedom to choose their own approach while providing essential, built-in tools like web searching, document handling, and local code execution. Whether you need human-in-the-loop feedback or modular skill extensions, this framework makes custom agent development truly intuitive. It is fully customizable, helping you build powerful agents that work with the model instead of against it. Start building your next generation of AI agents today.

🆔 @hackernewsgithubprojects

📦 sakanaai/sparser-faster-llms

Make Your LLMs Faster and Lighter with Sparsity

Are your language models struggling with high memory usage and slow speeds? This project introduces a breakthrough approach to making transformer models leaner and faster. By leveraging custom CUDA kernels built for H100 hardware, this repository provides the necessary tools to implement sparse training and inference using the efficient TwELL packing format. Whether you are running inference or training, these specialized kernels help you optimize your models for better throughput while reducing memory requirements significantly. It is a powerful way to push your AI performance to the next level. Start exploring these new sparse model capabilities today.

🆔 @hackernewsgithubprojects