📦 oliverz-dot/node-jepa

NodeJEPA Predicts Graph Secrets

Most graph learning tools either try to reconstruct raw data or force comparisons between different data views, which often leads to messy results. NodeJEPA takes a smarter path by predicting the hidden meaning of masked graph sections instead. It masks parts of a graph and asks the model to guess the underlying structure of those hidden parts based on the visible context. This approach skips the need for tricky data tricks or reconstruction steps. The result is a cleaner, more reliable way to understand complex networks. It is genuinely clever because it focuses on true prediction rather than simple copying.

🆔 @hackernewsgithubprojects

📦 zzaiyan/core-uir

CoRE-UIR: Fix Remote Sensing Images

Fix blurry or weather-damaged satellite photos in one go with CoRE-UIR. This tool cleans up images ruined by fog, rain, dust, or motion blur without needing separate tools for each problem. It works by first guessing what kind of damage is in the picture, then using that clue to guide the repair process. Think of it like a smart mechanic who knows exactly which wrench to use before even looking under the hood. The system handles single issues or messy combinations of bad weather all at once. It keeps the main image structure intact while swapping out the damaged parts. This makes restoring high-resolution aerial photos much faster and simpler.

🆔 @hackernewsgithubprojects

📦 sisyphbaous-dt-project/open-qingyi

Fix a Broken AI's Reading Comprehension

Transform a standard language model into a lightning-fast linear variant without losing its brain. This project replaces heavy attention layers with a lighter KDA architecture, cutting memory needs significantly. But here is the real magic: when the conversion broke the model’s ability to read multiple-choice questions, the authors didn’t just give up. They spotted a weird glitch where the AI stubbornly picked answer A every time, ignoring the actual content. By running a clever diagnostic that shuffled the options, they proved the model was confused by the format, not the knowledge. A targeted training fix restored its reasoning, turning a broken model into one that actually understands context.

🆔 @hackernewsgithubprojects

📦 avaxiao/retoken

ReToken: One Token for Better Video Retrieval

Make your vision language models hunt down specific details in long videos without getting lost. ReToken adds just one special learned token to existing AI models so they can instantly spot exactly which video frames contain the answer you need. Think of it like giving the model a tiny memory pin that marks the exact moment where the action happens. Instead of scanning through hours of footage, the model uses this single token to retrieve the crucial scenes directly. The training is surprisingly cheap, requiring only one token and one projection layer on a standard GPU.

🆔 @hackernewsgithubprojects

📦 7hhhhh/harness-g

Harness-G: Structured Search for AI Agents

Harness-G is the graph-based search engine that finally stops AI agents from getting stuck in retrieval loops. Traditional search agents often waste time fetching nearly identical results, failing to learn from diverse evidence. Harness-G fixes this by replacing open-ended queries with a finite menu of structured actions on a paragraph and entity graph. Instead of guessing search terms, the AI navigates a map of connected facts, allowing it to compare options and trace how early choices lead to later successes. This structure-aware credit system ensures the agent learns which specific steps actually unlock the right answers.

🆔 @hackernewsgithubprojects

📦 realzachi/shotluma

Shotluma: Editable App Store Screenshots with AI

Shotluma is the browser-based canvas tool that builds fully editable iOS App Store screenshots using artificial intelligence. Instead of generating static images, its agent places real text, shapes, and device frames you can select and change later. It works entirely locally, so your projects stay in your browser without any accounts or servers. You simply describe your app, upload your raw images, and the AI designs a complete multi-screen story for you. You can then tweak any element by hand or ask the AI to fix just one screen while keeping everything else intact.

🆔 @hackernewsgithubprojects

📦 calesthio/resonant

Resonant: Free Local AI Music Studio

Resonant is the free local AI music studio that lets you generate complete songs and mix professional tracks entirely on your Windows computer without sending data to the cloud. Instead of relying on expensive subscriptions, this tool gives you a full production environment where you can write lyrics, arrange instruments, and even play virtual keyboards to create music that stays safely on your own machine.

🆔 @hackernewsgithubprojects

📦 junrongh/controltile

Controltile: The Texture Tiler That Finally Gets It Right

Controltile is the texture tiling tool that finally gives you real control over repeating patterns without the usual messy artifacts. It uses a clever diffusion model to stitch seamless, high-quality textures from just a single reference image, solving the headache of awkward seams and repeating lines that plague traditional methods. Instead of just copying pixels, it understands the underlying structure of materials like brick or wood, generating new variations that feel authentic and grounded. You can tweak the style and structure independently, which is a game changer for game artists and designers who need consistent assets.

🆔 @hackernewsgithubprojects

📦 petergyang/human-review

Edit AI Files Visually with Human Review

Edit HTML and Markdown files directly in your browser and send your changes straight to an AI agent instead of typing out long, confusing instructions. You can click on any paragraph or image to leave comments just like in Google Docs, tweak basic formatting, or even drag and drop images to a new spot. The tool works right inside your favorite coding assistant, so you make visual tweaks in the browser and click one button to push every edit back to the agent for a quick refresh.

🆔 @hackernewsgithubprojects

📦 unclebob/swarm-forge

SwarmForge

SwarmForge is the tmux-based orchestration system that turns multiple AI coding agents into a reliable team of professional engineers. Instead of a single AI struggling alone, this tool coordinates several specialized agents like coders, cleaners, and architects, each working in its own dedicated workspace to avoid stepping on each other. The magic lies in a simple file-based handoff system. When one agent finishes its task, it drops a small file into a shared folder, and a background helper automatically notifies the next agent to pick it up.

🆔 @hackernewsgithubprojects

📦 unisic/unisic

Unisic: The Screenshot Tool That Does Everything

Draw arrows and shapes on your screen before the picture even snaps, then polish the result with powerful editing tools like blur and smart erasers, all without leaving the app. Unisic handles the entire workflow for Linux users by letting you capture regions, record short clips, extract text through OCR, and upload files to custom servers with a single click. It works beautifully on modern Linux desktops while respecting your privacy by collecting absolutely no data or tracking. The interface stays clean and fast, offering a refreshing alternative to basic screenshot apps that just hand you a raw image and walk away.

🆔 @hackernewsgithubprojects

📦 scroobius-pip/fudge-design-md

Fudge Design MD: Design Specs From Real Sites

Fudge Design MD is the design reference library that turns real website captures into detailed style guides. It solves the problem of reverse-engineering beautiful interfaces by automatically analyzing popular sites and extracting their actual color palettes, typography choices, spacing rules, and layout patterns. Instead of guessing how a site feels, this repository gives you plain text documentation that breaks down the visual language of hundreds of different projects. The most interesting part is how it translates those visual elements into practical guidance you can actually use, helping you understand why certain designs work so well.

🆔 @hackernewsgithubprojects

📦 inlitx/obsidian-dashboard-gallery

Obsidian Dashboard Gallery

Obsidian Dashboard Gallery gives you pre-built, beautiful home screens for your Obsidian vault without writing any code. You simply copy one of its four distinct dashboard designs into your vault and it instantly transforms your starting page into a functional hub for notes, tasks, and recent files. The Atlas dashboard is particularly clever because it uses JavaScript to pull in your bookmarks, track your editing activity with a heatmap, and even lets you capture quick thoughts right from the homepage.

🆔 @hackernewsgithubprojects

📦 get-bb/bb

bb: The Coding Agent That Builds Itself

bb is the coding agent IDE that writes its own plugins. Instead of a static tool, it features an internal agent that listens to your commands and automatically builds new extensions to handle complex tasks. You can steer conversations in real-time threads, and if the tool doesn’t exist yet, you simply ask the agent to create it. This means your development environment evolves as you work, turning manual setup into an automated conversation. It is a rare look at software that actually improves itself while you use it.

📰 https://news.ycombinator.com/item?id=49211169

🆔 @hackernewsgithubprojects

📦 fuxicodex/fuxi

FuXi: The Terminal AI Agent That Routes Your Code

FuXi is the developer terminal agent that finally connects any AI model to your workflow. It lives right in your command line, letting you chat with your code, edit files, and run commands without leaving the keyboard. The real magic is its smart routing system. Instead of being stuck with one provider, FuXi can juggle multiple AI models at once, automatically switching to a faster or cheaper option if your main one slows down or goes down. It even handles long conversations by keeping track of your project’s memory so you never lose context. You can use your own API keys or just log in for a managed experience.

🆔 @hackernewsgithubprojects

📦 remotion-dev/skills

Skills: The AI Tutor for Remotion

Skills is the AI tutor that finally teaches coding assistants how to build Remotion videos. Think of it as a cheat sheet for artificial intelligence, giving tools like Claude Code or Cursor a clear understanding of how to animate React components without guessing. Instead of struggling with confusing documentation, these agents receive structured best practices for everything from adding captions to rendering maps. It turns complex video editing into simple, reliable prompts. Your code becomes smarter overnight because the AI actually knows the right way to do things. Just install it and watch your development workflow improve instantly.

🆔 @hackernewsgithubprojects

📦 ssteelfactor-oss/kestrel

Kestrel: The AD Auditor That Sees Backdoors

Kestrel exposes the hidden backdoors that traditional Active Directory tools completely miss. While most security tools focus on mapping attack paths between users, this single executable digs into the directory attributes themselves to find shadow credentials, hidden objects, and dangerous certificate permissions that simply don't show up in a graph. It runs with zero dependencies and leaves a read-only footprint that defenders can actually track, telling you exactly what is vulnerable and how to fix it without ever modifying a single file. It bridges the gap between finding problems and fixing them by providing a prioritized list of critical issues with the exact commands needed to close the door.

📰 https://news.ycombinator.com/item?id=49093236

🆔 @hackernewsgithubprojects

📦 amap-ml/longhorizon-harness

LongHorizon Harness: AI that never loses track

LongHorizon Harness is the workflow engine that finally keeps AI agents on track during long, complex tasks. Most assistants lose context after hours of work, but this tool separates planning from doing by using three distinct roles. A manager holds the big picture, an executor focuses on the current step, and an auditor independently verifies every result before it is saved. This means the AI can switch between your desktop apps and the command line for hours without forgetting its goal. It runs on top of existing tools like Claude Code, ensuring that verified progress is never lost even if the system crashes.

🆔 @hackernewsgithubprojects

📦 rsmdt/the-startup

The Agentic Startup

The Agentic Startup lets a single AI tool act like a full engineering team. It forces you to write detailed specifications before writing any code. This prevents wasted effort by breaking down complex ideas into clear, testable steps. You get specialized roles for design, coding, and testing that work together. It turns a vague prompt into a structured plan. This helps builders ship better software without getting stuck in endless loops. You finally have a system that thinks before it types.

🆔 @hackernewsgithubprojects

📦 rafmacalaba/armada

Armada: Your AI Engineering Fleet

Armada is the development tool that turns any code repository into a self-organizing team of eight AI specialists. Instead of relying on a single bot that might mess up your work, this system sets up distinct roles for backend coding, frontend design, security auditing, and quality testing. You start by describing what you want, and a central orchestrator breaks that goal into clear steps. The specialists then work in parallel inside their own safe, isolated folders, ensuring your main code stays clean. Before any changes are shared, the quality team checks the work and verifies everything works as promised.

🆔 @hackernewsgithubprojects