📦 zjwzcx/awesome-astra-embodied-ai

How Astra Paints Real Things

Awesome Astra Embodied AI is the curated collection that tracks GPT-6 Astra controlling physical robots without traditional programming. Instead of writing complex code for every movement, developers simply describe a task like painting a bridge or grabbing a marker, and the AI figures out the physical motions on its own. The most striking example is watching a real robot arm hold a paintbrush and actually paint a picture of the Golden Gate Bridge using only a text prompt and camera feedback. This repository documents these breakthrough moments where language models bridge the gap between digital ideas and physical action.

🆔 @hackernewsgithubprojects

📦 ericlitman/open-pstack

Open Pstack: The AI Coding Agent That Actually Works

Open Pstack is the coding agent plugin that turns your AI assistant into a disciplined engineer. It is not a new model, but a set of rules and workflows that makes AI coding tools like Claude Code and Codex actually useful. It forces the agent to understand your code before changing it, test things for real, and show its work. If you use AI to help you code, this is how you get it to stop guessing and start shipping. It’s like having a senior dev who checks their own homework.

🆔 @hackernewsgithubprojects

📦 saya-ch/dsh-mobile

Talk to Your Phone UI

Control your computer's AI workspace directly from your pocket with DSH Mobile. This Android tool creates a secure bridge, letting you access the same real-time sessions and tools as your desktop without touching any source code. The coolest part? You can type a single command in your chat to instantly redesign the phone interface. Ask for a retro terminal look or a live system monitor, and the app refreshes with your custom layout in seconds. It handles the complex networking automatically, whether you are on local Wi-Fi or using a remote tunnel.

🆔 @hackernewsgithubprojects

📦 modelcontextprotocol/ext-skills

Ext Skills: The AI Manual That Ships Itself

Ext Skills is the working group that finally gives A.I. agents the step-by-step manual to go with their tools. Right now, servers hand over buttons, but agents often don't know how to press them in the right order. This project solves that by letting servers stream the actual how-to guides right through the same connection. Instead of installing static files that go out of date, the agent gets the latest instructions automatically. It means the knowledge stays in sync with the tools, and updates flow without you ever touching a terminal. Think of it as the product shipping its own user guide. If you build A.I.

🆔 @hackernewsgithubprojects

📦 nyckm/3d-camera-control-h3-minimax

Visualize 3D Camera Orbits for AI Video

Stop guessing how to make AI cameras move and start dragging a 3D camera in a visual editor. This ComfyUI plugin, called Camera H3, lets you orbit, zoom, and set keyframes with your mouse to plan a shot. The surprising part is how it works: instead of moving pixels, it translates your 3D movements into plain English text prompts. It writes a detailed description of the path and sends it to the MiniMax H3 model, which then tries to follow those instructions to generate your video. It is a bridge between spatial design and text-based generation. You draw the path, and it tells the AI exactly where to look.

🆔 @hackernewsgithubprojects

📦 flame-chasers/gtr

GTR: Text Search Without Labels

GTR is the computer vision framework that finds people using only text descriptions, completely skipping the need for human-annotated data. Traditional systems require thousands of hours for humans to manually label images with words, which is expensive and slow. GTR solves this by generating its own diverse, pseudo-labels to teach the model how to match text to images. It learns from these generated examples using a smart weighting system that ignores noisy mistakes, allowing it to become accurate without any manual intervention. This approach removes the biggest bottleneck in building searchable visual databases.

🆔 @hackernewsgithubprojects

📦 jzjzzzzzzz/agent-me

Agent-Me: The Transparent AI Twin

Agent-Me builds a digital version of you that you can actually see inside. Instead of a black-box chatbot, it uses a clear step-by-step process to find your notes, check the evidence, and verify the answer before it speaks. The most interesting part is that every step is inspectable, so you can see exactly why the AI said what it did. This makes it a real tool for understanding how AI represents your thoughts, not just a gimmick. If you want an AI that works with your context and stays transparent, this is the project to watch.

🆔 @hackernewsgithubprojects

📦 flyingwagner/scicode-verified

SciCode-Verified: The Benchmark Fix

Fix your AI benchmarks before they lie to you. SciCode-Verified repairs a famous scientific coding test that was accidentally rejecting correct code. The original benchmark used all-or-nothing scoring, meaning one tiny error in a frozen answer or a broken test would wipe out a perfect solution. The team audited every single problem, fixing hidden bugs and ambiguous instructions so the tests actually measure skill, not data glitches. This matters because the original version made top models look way worse than they are. By correcting these defects, we finally get a fair score. If you care about honest AI metrics, this is the version to use.

🆔 @hackernewsgithubprojects

📦 aryan-20-04/mdrap

MDRAP: The Market Data Guardian

MDRAP is the market data pipeline that refuses to let your data quietly disappear. It takes the chaotic, noisy feeds from multiple exchanges and cleans them into a single, verified stream without ever silently dropping a bad packet. The real magic is in its cryptographic audit trail. Every piece of data gets a digital fingerprint, creating an unbreakable chain of proof that shows exactly where it came from. If a regulator or a trader asks why a price was wrong, you can prove it in seconds. It is the difference between guessing and knowing. Stop trusting your data blindly; start proving it.

🆔 @hackernewsgithubprojects

📦 yangshun/ultrastorage

Ultra Storage

Ultra Storage is the storage library that lets your browser remember complex data like dates and maps without any manual setup. The standout feature is automatic key expiration. You can tell a stored item to vanish after a specific time, like sixty seconds. Once that deadline passes, the data simply acts as if it was never saved at all. This is incredibly useful for managing temporary sessions or access tokens without writing extra cleanup code. It handles the heavy lifting of tracking time and removing stale data for you, right on the client side. Your app stays clean, and sensitive information never lingers longer than it needs to.

📰 https://news.ycombinator.com/item?id=49693524

🆔 @hackernewsgithubprojects

📦 kunobi-ninja/kache

Stop Wasting Disk Space: Meet Kache

Kache turns your slow, disk-hungry builds into a lightning-fast, space-saving machine by storing every compiled piece of code based on its content, not its location. Instead of copying massive folders every time you switch branches or check out a new worktree, it uses hard links to point multiple projects at the exact same physical data on your disk. In a massive Firefox test, this meant a new worktree only added three gigabytes instead of writing out another full sixteen. It’s like sharing the same physical files across different folders without duplicating them. You get instant builds and a tiny footprint, proving that smart caching beats brute force every single time.

🆔 @hackernewsgithubprojects

📦 max99x/wutw-public

Worlds Upon The Wind

A fully playable, commercial-grade game is now sitting in plain sight for anyone to steal, and the developer just gave away the keys. Worlds Upon The Wind is a peaceful deckbuilder built on Godot where you resettle a shattered world, and this public release includes the entire production codebase, art assets, and build scripts. It is not a demo; it is a complete, tested reference for indie developers who want to know how a real game actually works under the hood. You get to see how the card system, map generation, and localization tools are wired together before the engine ever compiles. No paywalls, no secrets.

📰 https://news.ycombinator.com/item?id=49693784

🆔 @hackernewsgithubprojects

📦 zargarkhan1/quorum-alpha-dash

Thirteen A.I. Minds Argue Over Oil Prices

Instead of blindly following a single algorithm, a panel of thirteen distinct A.I. minds argues over West Texas Intermediate crude oil futures. Each agent attacks a specific angle, from geopolitical shocks to refinery margins, forcing the group to stress-test every assumption before reaching a verdict. The system is built to hunt for dissent, ensuring that final consensus only survives if it withstands rigorous internal conflict. It doesn't just spit out a trade signal; it shows you exactly why the committee believes the market is moving that way. The result is a transparent, human-in-the-loop dashboard that turns complex energy trading into a visible debate.

🆔 @hackernewsgithubprojects

📦 generelschwerz/llama.cpp

Run LLMs On Your Laptop With llama.cpp

llama.cpp is the C and C++ engine that lets you run large language models directly on your own laptop without paying for cloud servers or needing a massive graphics card. Most AI tools require expensive, specialized hardware to work, but this project strips everything down to plain code that runs efficiently on everyday computers. The surprising part is how small it makes models by compressing their math, allowing a standard laptop to handle complex text generation just like a data center. It supports everything from Apple Silicon to standard Intel chips, giving you full control over your AI.

🆔 @hackernewsgithubprojects

📦 pasta-devs/marinara-engine

Marinara Engine: Your Local AI Game Master

Marinara Engine runs a fully local AI game master that builds and plays complex roleplaying adventures directly on your own machine, keeping every private story and character safe on your hard drive. It solves the frustration of messy, cloud-dependent chat apps by giving you a free, plug-and-play engine that works immediately without any account creation. The software handles everything from managing character personalities to running live combat and weather effects, all inside one intuitive interface. It is an open-source toolkit for anyone who wants to build immersive AI worlds without sending their data to a distant server.

🆔 @hackernewsgithubprojects

📦 ldbumble/taskuary

Taskary: Your Inbox, Staffed by AI Agents

Taskary is the local-first task hub that turns your scattered work messages into organized, actionable tasks. Instead of letting emails, chats, and alerts drown you in noise, it funnels everything into one clear timeline. Then, AI triage reads each item, figures out what actually needs doing, and hands the work off to your favorite coding agents. You stay in control, reviewing and approving every step before anything runs. It’s like having a tireless assistant who sorts the chaos, does the busywork, and only pings you when a real decision is needed.

📰 https://news.ycombinator.com/item?id=49393717

🆔 @hackernewsgithubprojects

📦 fitchmultz/pi-posthorse

Posthorse: The Context Window That Keeps Your Coding Journey Alive

Posthorse is the context manager that keeps your coding journey alive without losing the plot. It solves a specific problem for users of the Pi coding agent: when a conversation gets too long, the model usually summarizes everything, which often leads to lost details. Posthorse takes a different approach. Instead of summarizing, it swaps the old context for a fresh one, like changing horses at a relay station. The full transcript stays safely stored in the background, but the model starts with a clean slate and a precise handoff of what matters. This means you can work on long projects without the model forgetting earlier decisions.

🆔 @hackernewsgithubprojects

📦 0xbakeer/qwen3.8-27b-4-bit-on-a-single-dgx-spark

Run Qwen 27B Fast on One Chip

A twenty-seven billion parameter model hits seventy-five tokens per second on a single DGX Spark chip. This repository, named Qwen 3.8 27B 4 bit on a single DGX Spark, is a measured proof that you do not need a server farm for this level of performance. It tests how squeezing the model into four-bit memory actually changes speed. The big surprise is that the speed boost only matters when serving one user at a time. Once you add a crowd, the advantage vanishes entirely. It also shows which settings break the model and which ones are safe.

🆔 @hackernewsgithubprojects

📦 tejasq/basically-ai-harness

basically-ai-harness: The Scaffolding That Makes AI Useful

basically-ai-harness is the minimal TypeScript tool that proves an AI model needs a structural cage to work. Built for a talk on harness engineering, it separates the raw model from the scaffolding that manages tools and memory. The key insight here is that guardrails catch structural errors, but only a verification step catches a confident lie. By giving one model a real browser and another a fake one, the repo shows that the harness is the moat. The model is rented, but the environment is yours.

🆔 @hackernewsgithubprojects

📦 franksuperg/electrochemical-battery-model-atlas

The Battery Model Atlas Explained

Sixteen out of nineteen complex battery simulations actually run without breaking. That is the core promise of the Electrochemical Battery Model Atlas. It is not a simulator itself, but a verified map of existing tools. Instead of guessing which codebase works, you get a checklist that proves which models run on your computer. It saves you weeks of debugging broken dependencies by showing you exactly what runs and what does not. You get a clear path from theory to working code. This is the fastest way to trust your simulation results.

🆔 @hackernewsgithubprojects