📦 spinabot/brigade

Brigade: Your Personal AI HQ

Brigade is the personal intelligence hub that actually connects your chat apps to real tools. Instead of just chatting, it executes tasks across your life, sending messages on Discord or Slack while managing your notes in Obsidian or Trello. The coolest part is the skill system, which lets the agent learn to use specific apps like Spotify or Apple Notes, turning it into a true digital assistant. It handles the messy backend work so you can just talk to it from any messaging platform. If you want an AI that does things instead of just saying words, Brigade is the one to try.

🆔 @hackernewsgithubprojects

📦 tokenrhythm/opensquilla

OpenSquilla: Smarter AI on Less Budget

OpenSquilla is the AI agent that packs more intelligence into every single token you spend. It works like a smart editor that cuts out all the fluff, keeping the core meaning of your conversation without the bloated word count. This means you get the same deep, thoughtful answers from your AI, but you pay for a fraction of the data, saving you real money on every chat. It is less about having a bigger brain and more about making the information you already have work harder.

🆔 @hackernewsgithubprojects

📦 xikhar/spiderbench

AI Built a Full Web-Swinging Game?

Spiderbench is the browser game that proves AI can build a full 3D world. It features a procedurally generated Manhattan-style city where you can web-swing, wall-run, and zip between skyscrapers. The entire project, including the character models, textures, and complex lighting, was generated by Claude under human direction. It’s not just a simple demo; it’s a playable open-world experience running entirely in your web browser. The coolest part is seeing a massive, detailed city with traffic and day-night cycles, all built from code. It’s a fascinating look at what large language models can do for game development. Check it out to see the future of AI-assisted creative coding in action.

🆔 @hackernewsgithubprojects

📦 togg53192-cmd/jailbreaks

Jailbreaks: The Prompt Collection

The jailbreaks repository is the prompt library that finally maps the hidden backdoors of major AI models. It is a collection of text instructions designed to override the standard safety filters built into large language models like DeepSeek and Kimi. Essentially, it gives the AI a new set of operating rules, telling it to treat your request as a simple task rather than a dangerous one. This lets the model drop its usual defensive hesitation and give you the full, unfiltered answer you asked for. It is a fascinating look at how much power you can hold with just the right words.

🆔 @hackernewsgithubprojects

📦 reladraw/reladraw

Reladraw

Most diagram tools force you to choose between two frustrating extremes: either you use a language like Mermaid and let an algorithm mess up your layout, or you waste hours clicking and dragging boxes in a visual editor. Reladraw is a clever new text-to-diagram language that perfectly bridges this gap by letting you define exactly where things go using relative, human-spoken language. Instead of fighting coordinate systems or automated layouts, you simply write that your database node is right of your app node and level with it. It calculates the exact layout automatically while respecting your precise design intent. It is the perfect middle ground for quick, gorgeous charts.

📰 https://news.ycombinator.com/item?id=49858513

🆔 @hackernewsgithubprojects

📦 sting8k/pi-vcc

How pi-vcc saves your AI coding memory

pi-vcc compresses your AI coding sessions into a tight summary without asking a language model to do the work. Instead of guessing, it uses simple rules to pull out key goals, file changes, and code commits, creating a structured note that shrinks your conversation by up to ninety-nine percent. The best part is that nothing is actually lost. A built-in search tool lets the agent dig back into the original, full history whenever it needs a forgotten detail. This keeps your AI assistant focused and fast, while ensuring no important context disappears into the void.

🆔 @hackernewsgithubprojects

📦 oldcircle/geo-sleuth

Geo Sleuth: Finding Photo Locations With AI

Pinpointing exactly where a photo was taken usually requires visible landmarks or readable text. Geo Sleuth changes that by using pure geometry to find your camera position, even if the picture is just a bridge and a mountain. The system analyzes the spacing of bridge piers as a distance ruler and matches the mountain’s silhouette against elevation data. It filters thousands of candidate spots until only one remains, showing you the exact spot and the direction you were facing. It works inside your favorite coding agents like Claude or Cursor. So next time you snap a picture, you can actually see how the computer solves the puzzle.

🆔 @hackernewsgithubprojects

📦 xddxdd/nix-cachyos-kernel

Nix CachyOS Kernel

Nix CachyOS Kernel automatically rebuilds and caches the CachyOS Linux kernel whenever the upstream project releases a new version, so NixOS users get performance-optimized drivers without waiting for slow, local compilation. A daily automated check watches for changes in the main CachyOS repository and instantly syncs this project to match. This means you get the latest hardware optimizations and security fixes directly through the standard package manager. It removes the manual work of tracking down specific kernel versions and lets your system stay current with minimal effort. You just add a single line to your configuration file, and your machine runs on the newest, tuned kernel available.

🆔 @hackernewsgithubprojects

📦 furongjun-1999/dsh-memory

Dsh Memory: The White Box AI That Actually Thinks

Dsh Memory is the white box artificial intelligence architecture that makes every thought and decision fully visible and auditable. Instead of relying on a black box that just spits out answers, this project builds a system where the machine maintains a clear world model and constantly checks its own logic. It is a fascinating look at how we can build AI that learns continuously while keeping strict trust guardrails so we always know why it decided something. If you want to understand how true machine intelligence might work without the mystery, this is the blueprint to watch. It is a bold step toward AI we can actually trust and understand.

🆔 @hackernewsgithubprojects

📦 avinash201199/system-design-kit

System Design Kit: The Blueprint for Massive Apps

One single file acts as a complete map for building internet giants like YouTube or Uber, a surprising feature hidden in plain sight. This repository, System Design Kit, solves the problem of information overload by gathering scattered tutorials, books, and real-world case studies into one organized guide. It walks you from basic networking concepts to complex database strategies, showing exactly how engineers handle millions of users without crashing. If you want to build software that actually scales, this is your shortcut to understanding the architecture behind modern apps.

🆔 @hackernewsgithubprojects

📦 yb2460/harness-anything

Harness Anything: AI Controls Your Office Apps

Harness Anything lets an AI agent operate closed-source programs like WPS Office, Microsoft Office, and Photoshop directly from the command line. It solves a major gap where artificial intelligence could only easily control open-source tools, but now it can drive real industry software using Windows automation. The most surprising part is that an AI can build complex PowerPoint slides, edit Photoshop layers, and manage Zotero libraries without a human ever clicking the mouse. This makes it incredibly useful for automating repetitive design and office tasks that used to require manual steps. If you are interested in letting computers handle the busywork for you, this is a great place to start.

🆔 @hackernewsgithubprojects

📦 huskerminion/techo5

Techo5: Your Echo Show 5, but it is actually Linux

Amazon’s Echo Show 5 was designed to be a locked-in gateway to their cloud, but Techo5 rips the Android guts out and swaps them for a tiny Alpine Linux system that runs a single, lean program. This open-source firmware kills Alexa entirely, so your voice commands go straight to your own Home Assistant server over a secure, local connection instead of leaving your house. The screen transforms from a wall of ads into a practical dashboard for your smart home, and the device even wakes itself up by listening locally without needing the internet.

📰 https://news.ycombinator.com/item?id=49869865

🆔 @hackernewsgithubprojects

📦 ahmedkhaleel2004/blitztree

BlitzTree Scans 3.6 Million Files in 14 Seconds

BlitzTree can scan three and a half million files on a Mac in fourteen seconds. Most disk tools take minutes to find large folders, but this app uses a Rust engine to read your drive in parallel, making it nearly six times faster than the built-in command line tool. You see a colorful map of your storage that lets you zoom into folders or drag items to the trash. It even spots safe caches and build files to delete with one click. No data leaves your computer unless you choose to use an AI helper. If your Mac feels slow, this is how you find the clutter without waiting forever.

🆔 @hackernewsgithubprojects

📦 kaankiziltug/logo-design-skill

Logo Design Skill: The AI Logo Designer

The most striking feature of Logo Design Skill is its private library containing over fourteen hundred real world logo files that an AI agent can search to understand design trends without copying existing brands. This project turns coding assistants into professional identity designers by guiding them through discovery briefs to final vector files. It uses simple Python scripts to automatically test every draft for readability at tiny sizes and color contrast issues before you even see the result. The system stops at a checkpoint to show you a concept overview with a recommendation, ensuring the final brand kit only matches a direction you actually approve.

🆔 @hackernewsgithubprojects

📦 angel291592/intent-router

Intent-Router: The Intent Compiler for AI Agents

A tiny prompt-only tool turns vague requests into strict contracts before any AI agent begins working. Intent-Router scans your actual code, docs, and ticket history to answer questions that already have answers, so it never interrupts you with forty-six follow-ups. Instead, it asks exactly one question that only you can answer, then saves the result as a machine-readable file called an IntentSpec. This means the next agent, session, or teammate starts from a shared contract rather than guessing from zero. It is the missing input layer that stops models from hallucinating context and forces them to respect what your project already says.

🆔 @hackernewsgithubprojects

📦 xiaopuouo/openchatx-mcp

OpenChatX MCP

Long-running artificial intelligence tasks can finally survive the end of a ChatGPT session without forcing you to manually reconstruct your work. OpenChatX MCP connects ChatGPT directly to your computer using the official, secure Model Context Protocol tunnel rather than unstable browser workarounds. By running a local runtime, it equips your regular chat with coding-grade file and shell tools, custom skills, and rules, all without draining your specialized agentic allowance. You can seamlessly bundle your folders into projects, delegate tasks to custom subagents, and generate unique session handoffs to transfer complete context to a fresh chat window instantly.

🆔 @hackernewsgithubprojects

📦 yuta-fywoo/jev-gui-delegate

Jev GUI Delegate

Running complex browser and desktop tasks with artificial intelligence can quickly drain your tokens because models keep reading the screen, selecting buttons, and repeating commands. Jev-gui-delegate solves this by introducing a local controller that handles the repetitive, deterministic clicks on Windows and Chrome, calling on a smart model only when semantic judgment is needed. In real-world tests, this smart delegation slashed the main model's token consumption by up to ninety-one percent and made end-to-end task execution over three times faster. It keeps operations safe with built-in user takeovers, strict boundaries, and outcome verification, saving computing power while keeping you in control.

🆔 @hackernewsgithubprojects

📦 lemomo-ai/lemo-opuscar

lemo-opuscar

An artificial intelligence agent can now direct, draw, score, and mix an entire short film from scratch using absolutely no video generation models or stock footage. The lemo-opuscar repository lets your coding agent act as a miniature film studio, rendering canvas and WebGL pages frame by frame to produce stunning animations. You just pick from thirty-nine distinct artistic styles, like retro eighty-column CRT terminal art, eighty-era cel anime, or traditional Chinese ink washes, and let the agent write the code to bring your story to life. It is a brilliant, hands-on way to watch code become cinema.

🆔 @hackernewsgithubprojects

📦 microsoft/windowsdeveloperconfig

Windows Developer Config: One Command Setup

Windows Developer Config is the setup tool that turns a fresh Windows machine into a complete dev station in one go. You run a single command, it installs your favorite tools like Git and VS Code, tweaks your settings, and even handles the tricky restart needed for WSL. It picks back up automatically after you sign back in. This is great if you hate spending hours configuring your environment. It gives you a standard, tested setup that is safe to run again. You get a clean, distraction-free workstation without the guesswork. It’s a huge time saver for anyone starting a new project or swapping laptops.

📰 https://news.ycombinator.com/item?id=49871976

🆔 @hackernewsgithubprojects

📦 supercorp-ai/supercov

Supercov: The QA Boss for AI Coder

Supercov turns your AI coding assistant into a disciplined engineer that proves its own work. Instead of just generating code, the agent runs your real test suite through Supercov to find exactly which lines and branches are missing coverage. It then writes one specific test for that gap, reruns the suite, and uses the diff to verify the improvement. This loop creates a verifiable record of quality, meaning your AI never claims success without solid evidence. It is a simple way to keep automated coding safe and accurate, ensuring every change is backed by real results. Stop trusting blindly and start demanding proof from your machine coder.

🆔 @hackernewsgithubprojects