📦 2929004360/ruoyi-qs-nvr

ruoyi-qs-nvr

Managing security cameras from different brands usually requires a mess of proprietary apps, but ruoyi-qs-nvr unifies over twenty different video protocols into one open-source platform. Built on a modern cloud architecture, this system connects directly to standard network streams, vehicle trackers, and official camera developer kits from major manufacturers. It solves the massive headache of managing fragmented hardware by giving you a single dashboard for multi-screen live previews, cloud recording, and remote camera steering. If you are building a custom surveillance network, this project offers a highly flexible starting point that brings all your feeds under one roof.

🆔 @hackernewsgithubprojects

📦 nekodificador/comfyui-nkd-klein-tools

comfyui-nkd-klein-tools

comfyui-nkd-klein-tools is the ComfyUI extension that finally turns messy Flux Klein image-generation pipelines into a clean, plug-and-play setup. Instead of drowning in a sea of custom wiring to build, blend, and detail your images, this project wraps everything into simple presampling and postsampling nodes. You just drop in your reference image, paint a mask, and write a prompt. It automatically handles the heavy lifting, like zooming in to repair faces with incredible detail, balancing your colors so edits match the original background, and even auto-detecting changed pixels so unchanged areas stay absolutely perfect.

🆔 @hackernewsgithubprojects

📦 mit-han-lab/kernel-design-agents

kernel-design-agents

AI coding agents can now automatically optimize ultra-fast graphics card code to win engineering competitions. A research group at MIT created kernel-design-agents, a workflow that guides Claude Code and other AI assistants to write, test, and polish high-performance CUDA computer code. Instead of guessing how to speed up hardware, the AI agent inspects your files, drafts a step-by-step strategy, runs real-world benchmarks, and records the performance data. It keeps iterating until it finds the fastest math formulas that actually work. If you want to see how intelligent agents can automate the hardest parts of hardware programming, check out this repository.

🆔 @hackernewsgithubprojects

📦 repowise-dev/claude-code-prompts

claude-code-prompts

Build smarter, more reliable AI coding assistants by using production-ready prompt templates designed for agentic tasks. The claude-code-prompts repository provides an independently authored collection of system, tool, and memory prompts inspired by the inner workings of Claude Code. It solves the common problem of AI agents getting confused, making destructive edits, or losing track of context during complex development tasks. The most interesting feature is its multi-agent coordinator prompt, which acts as a central brain to safely delegate tasks, synthesize findings, and run adversarial tests before making final code changes. Grab these templates to make your custom coding assistants instantly more disciplined.

🆔 @hackernewsgithubprojects

📦 iflytek/skillhub

SkillHub

Organize and share custom AI agent capabilities across your entire team without sending sensitive data to the cloud. This platform lets you run a completely private, self-hosted registry for agent skill packages directly on your own servers using Docker or Kubernetes. Instead of everyone writing duplicate code for different tools, developers can publish packaged skills under dedicated team namespaces, track versions, and install them instantly using a command-line interface. Built-in security scanners check for safety, and team admins can review packages before making them available globally. It is like having a private app store for your organization's AI helpers.

🆔 @hackernewsgithubprojects

📦 google/xrblocks

XR Blocks

Building immersive web-based mixed reality with smart artificial intelligence just got incredibly simple. The project called xrblocks makes it easy to construct complex browser-based spatial computing experiences by supplying a massive collection of plug-and-play components. Instead of starting from scratch, developers can instantly combine pre-made building blocks for real-time hand-gesture detection, audio spatialization, visual depth mapping, and Gemini-powered smart agents. It solves the headache of manual math and heavy setup by wrapping WebXR, 3D physics, and live language models into straightforward tools. You can assemble everything from interactive virtual math tutors to physical 3D ball pits in minutes. It is the perfect toolkit to easily bring spatial AI to life...

📰 https://news.ycombinator.com/item?id=48756638

🆔 @hackernewsgithubprojects

📦 amap-cvlab/abot-manipulation

abot-manipulation

Train robotic arms to perform smooth, precise tasks by predicting clean direct actions instead of filtering out messy mathematical noise. The abot-manipulation repository introduces a smart artificial intelligence model trained on over six million physical movements to master complex hand-eye coordination. Rather than relying on standard fuzzy algorithms, it uses an innovative action manifold learning technique to give real hardware incredibly stable and precise physical control. It is an incredibly clever setup for anyone wanting to build smarter, more reliable robot assistants that actually grasp objects correctly on the first try. Check it out and start teaching your robots to move with human-like precision today.

🆔 @hackernewsgithubprojects

📦 arpanghosh8453/open-dronelog

Open DroneLog

Your drone flight logs contain a goldmine of hidden data, and open-dronelog lets you unlock it entirely offline without a subscription. It is a high-performance, private dashboard that acts as a secure home base for all your flight logs, instantly parsing files from DJI and Litchi. Instead of sending your sensitive flight paths to third-party cloud servers, it processes everything locally on your own machine using a super-fast local database. You can replay your flights on interactive maps, track battery wear over time, and analyze detailed telemetry charts. It is the ultimate tool for keeping your flight data private and organized.

🆔 @hackernewsgithubprojects

📦 ifixai-ai/ifixai

iFixAi

Run 45 automated tests against your AI agent to catch hidden mistakes and policy violations before your users or regulators notice them. While standard dashboards show perfect metrics, agents can quietly leak permissions, fabricate claims, or fall for manipulative prompts. This clever diagnostic tool probes your system across core risk areas like fabrication, deception, and even frontier risks like sabotage, delivering a clear A-to-F letter grade in under five minutes. Because it runs independently from the model it is testing, you get a genuinely honest, reproducible audit of how your AI actually behaves in the real world.

🆔 @hackernewsgithubprojects

📦 quinnaho/claudemap

ClaudeMap

Claudemap is the visual mapping tool that finally helps you understand the code you build with AI. When you build software quickly using AI assistants, it is easy to lose track of how all the pieces actually fit together. Claudemap solves this by reading your repository and organizing your code into visual concepts that match how you actually think. You can zoom out to see the big picture, or zoom in to inspect details where colors highlight what is working and what is broken. It brings your codebase to life so you never lose the thread of your own creation.

🆔 @hackernewsgithubprojects

📦 wuyi2121/scan-planner

Scan Planner

Scan-planner is the local navigation system that helps robotic dogs find their way across incredibly complex terrain over long distances. Instead of getting stuck or bumping into walls, this smart planner maps out safe, collision-free paths in real time as the robot walks. It acts like a highly advanced low-level guide, taking high-level route directions and translating them into immediate, safe footsteps. This means four-legged robots can easily handle tight mazes, dynamic obstacles, and rough paths. It serves as a reliable physical foundation for robots that need to explore buildings, follow voice commands, or navigate tricky environments completely on their own.

🆔 @hackernewsgithubprojects

📦 microsoft/flint-chart

Flint Chart

Flint-chart is a visualization compiler that finally lets AI agents generate gorgeous, production-ready charts without fighting low-level configuration details. Instead of drowning in verbose settings for margins, scales, and axes, an AI agent simply declares what the data fields mean using rich semantic types like price, temperature, or country. Flint-chart then automatically handles the layout, sizing, and color schemes, translating that simple spec into native code for popular rendering engines like Vega-Lite, ECharts, or Chart.js. It means agents can reliably design beautiful charts, and developers can swap rendering backends instantly. Give it a try to experience hassle-free automated charting.

📰 https://news.ycombinator.com/item?id=48756577

🆔 @hackernewsgithubprojects

📦 guanyilun/agent-sh

agent-sh

Turn your everyday terminal into an AI-powered workspace with agent-sh. Just type a single greater-than sign in your normal shell and you are instantly talking to an assistant that already understands your current directory, your last command, and whatever error just popped up. Everything you love about your setup, like your custom aliases and vim, still works perfectly. Even cooler, you can easily swap the AI brain behind the scenes to run different models while keeping the exact same terminal interface. It is a brilliant way to bring smart coding assistance directly into your existing workflow without any tedious setup.

🆔 @hackernewsgithubprojects

📦 worldbench/danceopd

DanceOPD

DanceOPD is the open-source training recipe that finally lets you combine text-to-image generation, local editing, and global style transformations into a single model without them constantly fighting each other. Usually, trying to teach an AI generator a new skill like photo editing ruins its ability to create clean images from scratch. This clever system solves the interference problem by treating each capability as its own velocity field, then letting a student model learn directly from those specialized teachers along its actual generation path. It is a brilliant way to build one master image generator that handles everything from background swaps to hyper-realistic styling while keeping its core skills perfectly intact.

🆔 @hackernewsgithubprojects

📦 tilellilab/atome-lm

Atome LM

atome-lm is the tiny language model that finally runs fully offline on a two-dollar microcontroller. By combining ternary weights that require zero floating-point math with a clever three-way routing system, it squeezes functional text generation into just twenty kilobytes of flash memory. The project provides complete training code and a C99 inference engine that guarantees perfect, bit-exact parity between your Python training environment and the physical microchip. It is an amazing open kit for developers looking to put smart, domain-specific text generation onto ultra-low-cost, battery-powered hardware. Give it a spin to see how small language models can truly get!

🆔 @hackernewsgithubprojects

📦 kmx415/meshpoint

Meshpoint

Meshpoint is the open-source base station software that finally turns cheap, surplus Helium miners into ultimate Meshtastic powerhouses. By pairing a Raspberry Pi with a multi-channel concentrator, this clever setup lets you monitor and transmit across eight channels simultaneously instead of relying on a tiny, single-channel pocket radio. You get a gorgeous local web dashboard to chat, manage direct messages, and track every node in range on a live map. It even acts as an identity-preserving smart relay to help your local network grow. Just flash your Pi, load the dashboard, and dominate the airwaves.

🆔 @hackernewsgithubprojects

📦 espressif/esp-claw

ESP-Claw

Espressif has built an artificial intelligence agent framework that runs entirely on a tiny microchip costing just a few dollars. The project is called esp-claw, and it brings active decision-making directly to edge hardware like the ESP32. Instead of relying on rigid, pre-programmed code, this framework allows you to define and change your hardware's behavior simply by chatting with it through messaging apps like Telegram. It translates your text instructions into Lua scripts, downloads them, and runs them locally on the device in milliseconds. It turns simple silicon into a smart, self-programming assistant that keeps your private data off the cloud.

🆔 @hackernewsgithubprojects

📦 facebookresearch/autopartgen

AutoPartGen

You can now automatically break down a single flat image or a solid 3D mesh into a complete set of individual, fully functional 3D part meshes. The open source project autopartgen makes this possible by using an autoregressive approach to predict and discover the distinct components of an object. Instead of treating a 3D model as one solid block, it segmentally generates watertight individual parts like the petals of a flower or the limbs of a robot. It is an incredibly clever way to turn static shapes into ready to animate assets. Check out the repository to start decomposing your own 3D designs today.

🆔 @hackernewsgithubprojects

📦 evolvinglmms-lab/lmms-lab-writer

lmms-lab-writer

Write academic papers with AI without ever leaking your drafts or data. The lmms-lab-writer repository provides a local-first desktop LaTeX editor that puts privacy and productivity first. It automatically configures a lightweight LaTeX setup on your machine, eliminating painful background installations. You can chat directly with your choice of artificial intelligence models to write sections, edit text, and organize references in real-time. It also includes full offline capabilities and built-in Git tools for painless version tracking. It is a brilliant way to write beautiful research papers with smart assistance while keeping complete control over your intellectual property.

🆔 @hackernewsgithubprojects

📦 andymik90/aperant

Aperant

Hand over your entire coding to-do list to a team of digital developers and watch them build your software while you drink coffee. This desktop app called aperant connects directly to your git repository to coordinate multiple autonomous agents that work together. Instead of copy-pasting code snippets back and forth, you just describe the task, and the system automatically plans the architecture, writes the code, and runs tests to validate everything. It turns your development process into a visual board where you can watch tasks move from planning to completed. Grab the desktop app and let your new digital team handle the heavy lifting.

🆔 @hackernewsgithubprojects