📦 brinootk/ps5x360

Play Xbox 360 on PS5

PS5X360 is the experimental emulator that lets you play Xbox 360 games directly on your PlayStation 5. It takes the popular Xenia engine and rewrites it to speak the console's native language, meaning your DualSense controller buttons and audio output work exactly as expected, without any messy translation layer. It is not a finished product, and you will still see crashes, but the fact that real games like Sonic run on the hardware is genuinely wild. If you love retro gaming and enjoy tinkering with homebrew, this project is a fascinating look at how far cross-console emulation can go.

🆔 @hackernewsgithubprojects

📦 noahdunnagan/mcopt

mcopt: Minecraft's New Metal Renderer

mcopt is the Minecraft mod that finally lets your Apple Silicon Mac use its native graphics power instead of translating everything through an older, slower layer. It hooks directly into Metal, Apple’s graphics system, so the game renders far more efficiently than before. On a smaller MacBook with an A18 Pro chip, it pushes the frame rate up by nearly a hundred frames per second compared to the previous version, making flight and spinning tests feel noticeably smoother. It sits on top of the popular Sodium performance mod, automatically tuning itself to your specific Mac hardware.

🆔 @hackernewsgithubprojects

📦 zacharyzhang-ny/kigi-cli

Kigi: The Terminal Agent That Thinks in Graphs

Kigi is the terminal coding agent that finally works like a team, not just a single thread. Most AI tools just think, act, and repeat one task at a time, which gets slow for big projects. Kigi changes this by breaking your goal into a map of separate, parallel tasks that run simultaneously in the background. The coolest part is that it forces a hostile AI to try and break each piece of code before it is allowed to merge back into your project. It is like having a quality control gate built right into the workflow.

🆔 @hackernewsgithubprojects

📦 daseinlabs/parsec

Parsec cuts your coding agent bills by 39%

Parsec is the local proxy that slashes your coding agent costs by thirty-nine percent. It works by quietly trimming the massive pile of code and notes your AI agent gathers during a task, removing only the parts the model is unlikely to need. This simple cut means your agent solves more problems faster while spending far less, all without ever touching your actual prompts or changing how your code looks. Everything runs right on your machine, so your data stays private while you save real money. If you are tired of burning through tokens just to get a few fixes done, this is the efficiency boost your workflow actually needs.

🆔 @hackernewsgithubprojects

📦 airbytehq/airbyte-agent-sdk

Airbyte Agent SDK

Hand an AI agent a key that actually works. The Airbyte Agent SDK gives your code reliable, permission-aware access to over fifty major business tools, like Stripe, Slack, or Jira, without the messy guesswork of raw API calls. Instead of letting a language model hallucinate endpoints, this framework wraps every action in strict type safety and clear documentation, so your agent knows exactly what it can do and how to do it. It handles the boring parts, like retrying failed requests and managing credentials, so you can focus on building the agent logic rather than debugging connection errors.

🆔 @hackernewsgithubprojects

📦 colibri-social/colibri.social

Colibri: Own Your Chat

Spin up your own private chat space where you own every single message, thanks to Colibri. This open source platform runs on the A T Protocol, a decentralized system that means your conversations live on servers you choose rather than being locked away by a single corporation. It handles the heavy lifting of connecting people, managing members, and syncing text, voice, and forum channels so you can focus on the conversation, not the infrastructure. The most interesting part is that it runs on your phone, your desktop, and your tablet, giving you a unified experience across all your devices without the usual cloud lock in.

🆔 @hackernewsgithubprojects

📦 alexellis/glm-5.3-flash-4x-dgx-spark-switchless

Serve a 320B Model Without a Network Switch

Take a massive, three-hundred-and-twenty-billion-parameter AI model and run it on four affordable desktop workstations, completely skipping the expensive network switch that usually ties them together. This repository gives you the exact recipe to link those four machines into a closed loop of direct cables, turning them into a single, powerful brain that serves real coding tasks at impressive speeds. It solves the annoying problem of hardware procurement by proving you don't need that pricey rack switch for small clusters, saving you both money and wait time. Instead of buying a switch, you just wire the boxes together in a ring and let the software handle the rest.

📰 https://news.ycombinator.com/item?id=49508834

🆔 @hackernewsgithubprojects

📦 allenv0/scm

SCM Finds the Exact Second

SCM is the local search app that actually finds the specific second in your video library. You know that feeling when you remember a scene from a home video but can’t find the file? This solves that. It watches your folders and uses artificial intelligence on your Mac to understand what is happening in every frame. You can search for a moment, like a sunset, and it jumps you straight to that shot, not just the video file. It even finds spoken words and text on screen without sending your data to the cloud. It is private, fast, and incredibly useful for anyone who loves their digital memories.

📰 https://news.ycombinator.com/item?id=49952111

🆔 @hackernewsgithubprojects

📦 yoloyash/overtchat

OvertChat: Local AI for the Whole House

You can give your entire family their own private AI chat app, running entirely on your home server, with zero monthly fees. OvertChat is a self-hosted chat platform that connects to the AI models you already run locally, so your data never leaves your network. It ships with a one-click installer that sets up voice and web search right out of the box, meaning no confusing configuration files. The best part is that it works natively on Android, iOS, and desktop, so everyone in the house gets their own personal space to chat, save projects, and talk by voice.

🆔 @hackernewsgithubprojects

📦 kingkongrobotics/jumper

Jumper: The Crab Robot That Designs Itself

Jumper is the open-source crab robot project that turns plain language into physical motion. Most robotics code is locked in complex layers of software, but this repository lets you describe a movement or a skin in a single sentence to an AI assistant. It then builds the necessary simulation, trains the robot to walk, dance, or grip, and packages that behavior into a file ready for the hardware. This removes the steep learning curve between designing an idea and actually running it on a machine. You get a complete toolkit where your words become the blueprint for a real, moving creature.

🆔 @hackernewsgithubprojects

📦 dualface/kander

Kander: The Kanban Board for AI Agents

Kander drives nearly a thousand real production tasks by orchestrating multiple A I agents on a single terminal kanban board. This tool solves the chaos of parallel coding by physically isolating each agent in its own workspace and running a strict, independent review gate before any code is accepted. It acts like a rigorous project manager, catching logical bugs that cheaper models might miss and ensuring that automated delivery is actually reliable. The result is a controlled, parallel workflow where you can manage complex software development without constantly babysitting every single task, proving that structured oversight makes automated coding genuinely useful.

🆔 @hackernewsgithubprojects

📦 tier-flow/loopvl

LoopVL: The Tiny AI That Thinks in Loops

LoopVL lets you build a vision model that actually looks at your images multiple times before answering. Instead of the usual one-pass approach, this one billion parameter model reuses the same brain layers over and over, looping through the visual data to refine its understanding. It is surprisingly good at complex reasoning, solving math problems, and spotting details in photos. You can run this right on your own graphics card, which is rare for this level of performance. It is a clever way to get big results without needing a data center. If you want to see how much a little repetition can do, check it out.

🆔 @hackernewsgithubprojects

📦 donvito/agent-monitor

Agent Monitor

Watching your AI coding agents spawn a new subagent and animate it into a live graph is the coolest thing I have seen. Agent Monitor is a local dashboard that automatically finds your active Codex or Claude Code sessions and maps out the entire family tree of agents. It shows you exactly who spawned whom, what model each one uses, and how many tokens they are burning in real time. Because everything runs locally, your data stays on your machine, giving you total privacy. It is basically a mission control for your code helpers.

🆔 @hackernewsgithubprojects

📦 alejandroqh/browser39

Browser39: The Headless Browser That Saves 75x Tokens

Browser39 is the headless browser that returns exact page sections in just one hundred and ninety-six tokens instead of dumping the whole site. Built in Rust, it runs JavaScript without needing Chrome and turns messy web pages into clean, readable text for artificial intelligence. In a real test, while other tools burned thousands of tokens or missed content entirely, this tool grabbed the exact Wikipedia section needed at zero cost. It is a single, lightweight binary that keeps everything local, so your agent gets precise data without sending secrets to the cloud. If you are building tools that read the web, this is the efficient way to do it.

🆔 @hackernewsgithubprojects

📦 vavallee/bindery

Bindery: The Self-Healing Book Manager

Bindery automates the entire journey of acquiring and organizing digital books, handling the search, download, and filing without you lifting a finger. It solves a frustrating problem in self-hosted libraries: previous tools collapsed entirely when a single metadata website vanished, leaving users with broken, unusable apps. This project fixes that by pulling data from several independent, stable sources, so if one service goes offline, the others quietly step in to keep your catalog running. It also lets you drag in your old database with one command, instantly bringing all your authors and settings into a fresh, modern interface.

🆔 @hackernewsgithubprojects

📦 lotvulture/lotvulture

Lot Vulture

Lot Vulture is the parking system that turns existing security cameras into live occupancy maps without touching the pavement. Most smart lots require drilling into asphalt to install magnetic sensors for every single spot, which is expensive and breaks easily. This project skips that entirely by running computer vision directly on the camera feed to detect cars in real time. It works with standard IP cameras from any brand, meaning you get a live dashboard and instant alerts using hardware you probably already own.

📰 https://news.ycombinator.com/item?id=49550503

🆔 @hackernewsgithubprojects

📦 qualixar/superlocalmemory

Superlocalmemory: Your AI's Local Brain

Stop letting your AI forget everything after every chat. Superlocalmemory gives your AI agents a private, local brain that actually remembers what you told it, right on your own machine. It is not just a database, but a controlled space where your agents can store, find, and organize facts without sending sensitive data to some far-off cloud. The coolest part is that it learns from your team, sharing knowledge securely so everyone works smarter together. It is like giving your tools a long-term memory you control. If you build with AI, this is how you make them truly useful partners, not just parrots.

📰 https://news.ycombinator.com/item?id=49963368

🆔 @hackernewsgithubprojects

📦 ganyuanran/autoloom

Autoloom: AI Coding That Actually Thinks

Stop letting your code agent guess. Autoloom makes your AI think like an engineer before it writes a single line. It asks why you even need to change the code and checks if you're adding unnecessary complexity. Then, when it finishes, it proves its work with actual evidence instead of just saying it's done. This is a free desktop app for Windows that plugs in any model you want. It turns chaotic AI coding into a disciplined process you can actually trust, saving you from endless debugging. Your AI stops hallucinating and starts delivering real, verified software. Give it a spin and watch your code quality shoot up.

🆔 @hackernewsgithubprojects

📦 cneuralnetwork/kharcha

Kharcha: Your bank SMS ledger

Kharcha reads the bank text messages already sitting in your phone to build a private spending ledger. It pulls out the amounts and merchants, keeping the original message right next to the entry so you can always double-check the facts. The whole thing lives locally on your device, meaning no account sign-up and no data leaving your phone unless you explicitly turn on an encrypted backup. It is a clever, low-maintenance way to finally stop guessing where your money went. If you want a clearer picture of your daily spend without handing your finances to a giant cloud app, this is the tool to try.

🆔 @hackernewsgithubprojects

📦 wuyoscar/aisafetyhot-hub

AI Safety HOT: Daily AI Safety Digest

Keep your AI agent up to date on safety by connecting it to a live daily digest of attacks, defenses, and governance news. The AI Safety HOT Hub publishes a curated Chinese briefing every morning, translating complex academic papers into plain language and providing open data in standard formats for your research. The standout feature is its agent integration, allowing your automated tools to query this live security intelligence directly without needing any API keys. You no longer have to manually track scattered research to stay safe. This project turns noisy technical updates into a single, actionable daily brief for any developer building with large language models.

🆔 @hackernewsgithubprojects