📦 pxticks/h3-motion

H3 Motion: Turn Photos Into Moving Videos

H3 Motion lets you take a single still photo, pick out any object inside it, and turn that object into a moving, animated clip with sound. It is a local web app that connects to ComfyUI and uses open models to find subjects, remove them cleanly from the background, and then animate them along paths you draw yourself. You can also mask parts of existing videos and change what happens inside that mask. It is built for people who want to experiment with AI video tools without relying on closed cloud services. If you are curious about how local AI video editing works, this is a good place to start.

🆔 @hackernewsgithubprojects

📦 deepseek-ai/dsh-superagent-prototype

dsh-superagent: The Coding Agent That Only Speaks When It Matters

DeepSeek’s research prototype, dsh-superagent, quietly watches your GitHub repository and waits for a specific, allowed request before it starts coding. Instead of constantly interrupting your team with notifications, it operates on a strict rule of giving value without demanding attention. When a trusted person asks for work, the agent spins up its own isolated environment to write the code and submit a pull request, staying completely silent until the work is finished. It only breaks its silence to deliver that final result or ask one specific question.

📰 https://news.ycombinator.com/item?id=50316116

🆔 @hackernewsgithubprojects

📦 khalilgharbaoui/opencode-claude-code-plugin

Run Claude Code Inside OpenCode

Plug Claude Code directly into OpenCode without touching your login. This tool spawns the official CLI as a background process, letting your editor inherit your existing subscription or API key safely. It solves the headache of manual proxying by keeping sensitive credentials locked inside the official client. You get the best of both worlds: Claude’s powerful reasoning with OpenCode’s flexible interface. It also lets your editor execute dangerous commands like file edits under its own permission system. So you get Claude’s brain with your editor’s safety guardrails. No more copying tokens or breaking terms of service. Just one line in your config and you are running.

📰 https://news.ycombinator.com/item?id=50318664

🆔 @hackernewsgithubprojects

📦 ashayas/icloud-s3

iCloudS3: The Free S3 Server

iCloudS3 is the object store that hides your files inside photo pixels. It is a Rust server that speaks the Amazon S3 protocol but actually saves your data inside lossless PNG images uploaded to iCloud Shared Photo Albums. Apple's policy says these albums do not count against your five gigabyte storage limit, so you can technically store terabytes of data for free. The project encodes file bytes into the colors of the images, meaning your database backups literally live as pictures in your photo app. It even includes a benchmark showing it is hundreds of times faster than Cloudflare R2 because it just writes to your local disk.

🆔 @hackernewsgithubprojects

📦 hyz317/hairdeck

One Photo to 3D Animated Hair: HairDeck

Take a single flat photo of a character and turn it into a fully animated, three-dimensional hairstyle. Hair Deck is an open-source tool that solves the tedious problem of manually modeling hair for digital avatars. Instead of creating a static mesh, it builds individual, flexible hair cards and automatically attaches spring-bone chains, allowing the hair to swing and bounce naturally when the character moves. You can then export the result directly into popular animation formats like VRM or GLB. It is a powerful way to bring static images to life with realistic motion, making it incredibly useful for anyone creating virtual characters for games, social media, or VR applications.

🆔 @hackernewsgithubprojects

📦 sii-liulab/manimux

ManiMux: One Command for Real Robot Experiments

Drive a physical robot arm with a single terminal command. ManiMux is an open-source tool that handles the messy job of connecting AI models to real hardware. It starts the cameras, the brain, and the control software all at once, keeping everything synchronized in one clean interface. If you press the stop key, it shuts down safely, protecting the machine. This removes the headache of manual setup for robotics researchers. You get to focus on the experiment, not the wiring. Try it out and see how easy real-world automation can be.

🆔 @hackernewsgithubprojects

📦 phuryn/product-engineering

Product Engineering

Ship a real software product with coding agents without writing a single line of code yourself. This collection gives non-technical founders and product managers a complete playbook to take an idea all the way to paying users. It connects multiple A I agents to handle the heavy lifting, from building features to running security checks. You act as the director, reviewing prototypes and approving releases while the agents do the technical work. It turns the chaotic process of building into a clear, manageable path. If you want to launch without hiring an engineering team, this is your starting point.

🆔 @hackernewsgithubprojects

📦 qeude/defi

Defi: The macOS Window Manager That Scrolls

Defi is the window manager that turns your Mac desktop into a single, scrollable strip. Instead of stacking windows into tiny, cluttered corners, it arranges them in columns that flow horizontally. Open a new app, and the strip grows; move to a window beyond the edge of your screen, and the desktop scrolls to bring it into view. It is built natively for the latest macOS, so it feels fast and smooth, much like a modern mobile interface. You can still use your mouse and keyboard exactly as you always have. This is a clean, minimal way to keep your work organized without the usual window chaos.

🆔 @hackernewsgithubprojects

📦 maciekzygmunt/excaliself

Excaliself: Your Team's Self-Hosted Whiteboard

Turn your own server into a private, real-time whiteboard for your entire team. Excaliself takes the beloved Excalidraw editor and puts it on your infrastructure, meaning no external cloud dependencies and full ownership of your data. The coolest part is that it gives your AI tools direct access to these drawings. You can connect Claude or other smart agents to read, search, and even create new diagrams right inside your secure instance. It’s like having a smart assistant that knows exactly where every flowchart lives. Stop letting your ideas scatter across random websites. Grab your team’s work, keep it on your terms, and let your AI help organize it.

🆔 @hackernewsgithubprojects

📦 maharmstone/moxnox

Moxnox: The OS That Merges Windows and Unix

Moxnox is the hobby operating system that finally lets you run Unix software on a Windows NT-style kernel. It takes the internal architecture of Windows, including its file handling and system calls, and wraps it in a layer that speaks the language of Linux. This means you can compile standard Unix tools, like Git and Python, and run them natively without a compatibility layer. The most surprising part is that it is self-hosting, meaning you can build the entire operating system from within itself. It runs on common hardware and sits on a standard file system.

📰 https://news.ycombinator.com/item?id=50267569

🆔 @hackernewsgithubprojects

📦 clawboo/clawboo

Clawboo: The AI Team Manager

Clawboo is the open-source studio that puts a whole team of AI agents to work for you. It’s a self-hosted tool that lets you spin up a crew of agents, from marketing experts to coding specialists, all in one command. The coolest part is that it features a leader called Boo Zero, who automatically breaks down your request into tasks and assigns them to the right teammate. You can watch this entire process happen live on a visual board, seeing exactly who is doing what in real time. It’s like having a manager who delegates the busy work so you just get the finished result back.

🆔 @hackernewsgithubprojects

📦 anthropics/claude-code-playground

Claude Code Playground

Anthropic’s own developers share their private setups in a new public repository called the Claude Code Playground. This is not a product to install, but a transparent look at how the team actually configures their coding assistant. You get access to the exact prompt files, automation hooks, and settings they use daily, along with honest notes on what failed during the process. It solves the guesswork of learning by example, showing real workflows rather than polished tutorials. Every entry includes the specific model version and the steps to replicate the result. Think of it as a behind-the-scenes look at the tool’s power, built by the people who make it.

🆔 @hackernewsgithubprojects

📦 ghorbelhamdi/buddy-android

Buddy: Your Phone Is Now a Computer

Buddy turns your Android phone into a full computer by running its own tiny Linux system right inside the app, with no computer, cable, or root access required. This floating chat bubble lets AI agents like Claude Code physically use your phone by reading the screen, tapping buttons, and typing text to finish tasks for you. You just tell it what to do, like sending a text or buying something, and it handles the clicking while asking for your permission before any action that sends or spends money.

🆔 @hackernewsgithubprojects

📦 akdeb/open-toys

OpenToys: Local AI Toys with ESP32

OpenToys turns your Apple Silicon Mac into a private brain for physical AI companions. You plug an ESP32 board into any toy or robot, flash the firmware directly from the app, and let the Mac handle all the heavy lifting locally. No cloud, no subscriptions, and no data leaving your home. The most surprising part is that the device creates its own open Wi-Fi hotspot, letting your Mac connect directly to it without needing a router or internet access. It even supports voice cloning, so you can give your creation a unique sound.

🆔 @hackernewsgithubprojects

📦 ruc-nlpir/agent-reflex

Agent-Reflex: Teaching AI to Recover From Failures

Agent-Reflex teaches AI agents to recover from real-world mistakes like tool timeouts and user interruptions. It solves the problem of agents failing when deployed in noisy environments by learning reflection skills from failures. This is useful because it makes AI more reliable and robust. The one surprising thing is that it learns to notice and fix errors without needing explicit instructions. It uses a method called RSD to mine skills from failures and distill them into training data. The result is an agent that stays on track even when things go wrong. So the takeaway is that reflective learning is key to building dependable AI systems.

🆔 @hackernewsgithubprojects

📦 openrobokit/roboskills

RoboSkills: The cheap way to build robot agents

RoboSkills is the robotics toolkit that lets AI agents build complex robot behaviors for pennies. Instead of spending thousands of dollars on massive language model calls, this project uses specialized tools to keep costs down to fractions of a cent. The coolest part is the motion generation workflow, which takes a simple video and turns it into a full 3D animation of a robot hand performing the exact same action. It does this by reading the movement and applying it to a digital model, all without expensive physics calculations. You get a realistic demo for the price of a coffee.

🆔 @hackernewsgithubprojects

📦 ln-dev7/landing-pages

Landing Pages: The Portfolio That Breathes

Landing Pages is the open-source collection of websites that treats every project as a distinct piece of art. Each site is a standalone story, like a fictional paper flight bureau or a hand-delivered gift service, built with modern tools to feel alive. The standout feature is how each page uses 3D objects and motion to tell its unique brand story, rather than looking like a generic template. It is a masterclass in design, showing how code can create emotional, memorable experiences for any creative developer.

🆔 @hackernewsgithubprojects

📦 lensfamily/lenserase

LensErase: The Eraser That Understands Context

LensErase doesn't just delete objects from photos; it surgically removes them along with their shadows, reflections, and associated visual effects. This foundation model works by taking a simple image and a selection mask, then intelligently reconstructing the background to fill the gap. It handles complex scenarios like watermarks, text, and semi-transparent layers without you needing to type a single prompt. By training on over one hundred million paired examples, the model learns to preserve the surrounding content with high fidelity. The result is a clean, seamless edit that looks like the object was never there at all, making it a powerful tool for anyone needing precise, high-quality image cleanup.

🆔 @hackernewsgithubprojects

📦 fyzhang1/expactivator

ExpActivator: The Memory Trick

A frozen AI agent can look at your past actions and still pick the wrong button, but this tool fixes that. The project, called ExpActivator, argues that just finding relevant history isn't enough; the agent actually has to activate the right experience for the current moment. It works by matching your current screen to past situations in a hidden feature space, giving the model a reference action that it can then adopt, adapt, or reject entirely. This is a big deal because it requires no retraining. You don't touch the underlying model parameters.

🆔 @hackernewsgithubprojects

📦 vela-desktop/vela

Vela

Run a Linux desktop where every animation is locked to your monitor's actual refresh rate. Vela builds its own rendering engine to ensure that a slow application never holds back the rest of your screen. If one program stalls, the cursor and other windows keep moving smoothly on the next frame. This keeps the interface fluid even when a game or app lags. It is a complete, modern desktop experience designed for speed. Check it out to see how responsive your computer can really feel.

🆔 @hackernewsgithubprojects