📦 lnkiai/m3e-canvas

M3E Canvas: The Browser Sketchpad for AI Coding Prompts

Drag buttons, cards, and sliders onto a canvas to sketch out a full mobile app interface right in your browser. M3E Canvas is a design tool that lets you arrange Material Design components, link screens together, and preview the user flow before you write a single line of code. The magic happens when you are done sketching: the app translates your entire design into a concise, natural-language prompt. You copy that text, paste it into an AI coding tool, and the AI builds the actual application for you. It is a clever bridge between visual design and AI development, turning your rough ideas into working code without the usual back-and-forth.

🆔 @hackernewsgithubprojects

📦 macanalog/spicexplorer-release

SpiceXplorer Release: The Open Analog Design Workspace

Bring analog circuit design to life by spinning up a full simulation environment without installing a single library on your computer. SpiceXplorer from the MacAnalog research group gives you a Docker container that packs a real SPICE engine and three open process kits into one image. You can start exploring complex amplifier topologies or LDO stability immediately, right in your browser, with zero setup friction. It is the fastest way to see actual waveforms and performance metrics without wrestling with local dependencies. If you want to tinker with analog blocks without the usual install pain, this is your shortcut to a working lab.

🆔 @hackernewsgithubprojects

📦 boschresearch/memorydrivor

MemoryDrivoR: Driving Cars on Pure Memory

MemoryDrivoR is the self-driving research tool that proves you don't need live camera feeds to drive well. Instead of watching the road in real time, this model relies entirely on a memory of previous drives to predict safe paths. It is a clever way to test if current autonomous driving benchmarks are actually measuring real-time awareness or just pattern matching. By stripping away the need to observe live traffic, the project reveals just how much static map data and past experience can do. It is a fascinating diagnostic for the field, showing that high scores can come from memory alone rather than true perception.

🆔 @hackernewsgithubprojects

📦 chrisyrniu/isaaclab-decoupled-wbc

IsaacLab-Decoupled-WBC: Train Humanoid Robots on One GPU

IsaacLab-Decoupled-WBC trains a humanoid robot to walk and balance using just a single graphics card, which is honestly surprising for robotics work that usually demands a server rack. This framework handles the entire pipeline from training a teacher policy to distilling it into a smaller student policy that runs on actual hardware. It focuses on decoupled whole-body control, meaning the robot manages its legs and waist to stay stable while keeping its arms in fixed positions, a setup that was actually used for the Unitree G1 experiments. You can even drive the simulated robot with your keyboard to see how it responds to movement commands before deploying anything.

🆔 @hackernewsgithubprojects

📦 6lyc/recharness

How Recharness Evolves Recommendation Models

Recharness is the experimental framework that lets an AI actually improve its own recommendation code. Instead of a human manually tweaking settings, this system uses a smart guessing algorithm to pick where to look, while a language model writes the actual improvements. It tests these changes in safe, isolated sandboxes to see if they work. If a new version scores better on test data, it keeps it; if not, it throws it out and tries a different angle. This creates a self-correcting loop that finds better models faster than guessing randomly. It’s basically a tireless research assistant that iterates until it finds a winner, all within a strict time budget.

🆔 @hackernewsgithubprojects

📦 kaiwu5/awesome-ai4ai

Awesome AI4AI: The Map of Self-Improving AI

Twenty-two-three papers. That is the current count in a living catalog called Awesome AI4AI, which tracks the specific research question of whether artificial intelligence can reliably improve itself. This is not just a list of cool tech demos; it is a curated map of the field, organized into benchmarks, runtime design, and model architecture. The most surprising part is the evidence audit, which breaks down thirty-five real systems to see exactly where humans and machines take over. It turns out that while AI systems can handle execution and repair, the core goal is still owned by humans.

🆔 @hackernewsgithubprojects

📦 jonasbaeumer/herdr-file-annotator

How to review AI code changes in Herdr

The Herdr File Annotator lets an AI coding agent freeze its work and open a live review pane directly inside your terminal. This plugin solves the friction of trusting automated code changes by forcing a real human checkpoint. Instead of scrolling through endless chat logs, the agent guides you to specific code sections while you read the actual file changes. You can tag lines as fixes or questions, and that feedback flows straight back to the AI so it knows exactly what to correct.

🆔 @hackernewsgithubprojects

📦 doofzoff/simurg

Simurg: Real-Time LLM Guard

Simurg stops broken AI answers before you ever read them. When a language model starts looping or drifting into nonsense mid-sentence, this tool cuts the stream and triggers a fresh retry instantly. It works in three lines of code and runs on any standard model server. No GPU, no training, just plain Python. It watches your text as it flows, catching errors while they happen so bad data never reaches the screen. If your model is misbehaving, this is the safety net you need. Keep your users happy by filtering out the glitches automatically. Try it on your next local model deployment and see the difference in quality.

📰 https://news.ycombinator.com/item?id=49425784

🆔 @hackernewsgithubprojects

📦 microsoft/tgrep

Tgrep: The Grep That Actually Thinks

Find that one function in a massive codebase without waiting for your computer to burn out. Tgrep, made by Microsoft, skips the slow habit of scanning every single file just to find a match. Instead, it builds a smart map of your code ahead of time. While standard search tools re-read everything from scratch, this tool only looks at the specific files that could possibly contain your answer. You build the index once, keep the server running, and suddenly your search queries are nearly instant. On huge projects, it runs up to fifty times faster than the usual options.

📰 https://news.ycombinator.com/item?id=49562928

🆔 @hackernewsgithubprojects

📦 nklmilojevic/sofka

SoFka: The Kubernetes Tool That Tells You Why It's Broken

SoFka is the Kubernetes interface that finally tells you why your app is broken. Instead of forcing you to dig through logs and settings manually, it uses a built-in logic engine to pinpoint exactly what went wrong. It connects directly to your cluster without freezing your screen, so you can check multiple issues at once without waiting. If your application crashes, it instantly shows you the failing step and the specific error message, saving you hours of guessing. It is fast, lightweight, and designed to keep you in flow while you fix things.

📰 https://news.ycombinator.com/item?id=49359674

🆔 @hackernewsgithubprojects

📦 caura-ai/caura

How Caura Makes AI Teams Smarter

Caura gives your team of A.I. agents a shared brain that gets smarter every time they work together. Instead of each bot keeping secrets or repeating the same mistakes, they write down what they learn into one common space. The coolest part is that the system actually figures out if two agents are saying opposite things and automatically fixes the error. It is like a team that never forgets a lesson. Since it runs entirely on your own machines, you keep full control over what your digital workers know. If you are building a fleet of bots, this is the missing piece that makes them truly useful.

🆔 @hackernewsgithubprojects

📦 sublang-ai/cligent

Cligent: One Interface for Every AI Agent

Cligent turns the chaos of five different artificial intelligence coding tools into one simple TypeScript command. Instead of writing separate code for Claude Code, Codex, or Gemini, you plug in a single adapter and receive a unified stream of events, meaning your app never needs to know which specific model is actually working underneath. The most fascinating part is a built-in demonstration called Tmux Play, which splits your terminal so a lead agent can dispatch tasks to different models simultaneously, letting you watch multiple artificial intelligence workers stream their progress side-by-side in real time.

🆔 @hackernewsgithubprojects

📦 boardui/boardui

BoardUI: The Source Code Design System

BoardUI is the design system that gives you the actual code, not just a library. It drops real React files directly into your project, so you own every pixel without fighting a black-box dependency. The coolest part is that the homepage is a working AI chat app. You just paste your own API key, and it instantly starts streaming replies using your preferred model provider. It handles the tricky stuff like keeping your secret key safely on the server while the chat history lives right in the user's browser. This means you get a fully functional, customizable AI interface without building the plumbing from scratch.

📰 https://news.ycombinator.com/item?id=49559330

🆔 @hackernewsgithubprojects

📦 nomadoor/comfyui-panorama-stickers

ComfyUI Panorama Stickers: 360 Editing Made Simple

Place stickers directly onto a three hundred and sixty degree panorama, then extract a normal flat photo from any angle without leaving your workflow. ComfyUI Panorama Stickers handles the complex math of spherical projections so you can focus on your creative composition rather than wrestling with distorted geometry. The real magic is the cutout camera, which lets you frame a specific perspective inside the sphere and output it as a standard rectangular image, effectively turning a panoramic world into a shareable snapshot. This saves hours of manual cropping and fixing perspective warps in other software.

🆔 @hackernewsgithubprojects

📦 alayalab/evoke

Evoke: The AI That Remembers Where It Has Been

Evoke generates thirty seconds of coherent video without forgetting the background, a feat that usually breaks down as AI models lose track of their own scene. Most video generators rely on a sliding window that forgets earlier details, but Evoke stores the 3D geometry of the world in a separate memory bank. This allows the model to look back at previous moments to keep the scenery consistent, no matter how long the video runs. It also lets you change the prompt while the video is generating, so a calm lake can instantly turn into a stormy scene without any cuts or restarts.

🆔 @hackernewsgithubprojects

📦 prysai/prysai-llm-playbook

Prysai LLM Playbook: The Evidence-Led Guide

The Prysai LLM Playbook is the evidence-led field guide that finally separates real capability from hype. Instead of relying on vague theory, this project builds a complete, transferable manual for working with language models like ChatGPT, Claude, and Gemini. It treats AI as a professional tool, offering a structured curriculum that teaches you exactly how to design tasks, set strict boundaries, and verify results before you trust the output. This matters because it turns chaotic experimentation into a repeatable skill that scales from individual projects to entire teams. If you want to use AI with precision rather than guesswork, this is the blueprint you have been waiting for.

🆔 @hackernewsgithubprojects

📦 pingdotgg/t3code

T3 Code: Control Your Coding Agents From Your Phone

T3 Code is the open-source control surface that lets you command the coding agents already running on your computer. It solves the frustrating problem of being locked to your desk, letting you manage sessions from a dedicated mobile app, a web browser, or a desktop client. You can steer popular providers like Claude Code, Codex, and Cursor directly from your pocket, turning your machine into a remote workhorse. It is entirely free, meaning no extra subscription costs, and it gives you full visibility over what your tools are doing.

🆔 @hackernewsgithubprojects

📦 qwenlm/e-commercebench

E-Commerce Bench: The Ultimate LLM Business Test

E-Commerce Bench is the long-horizon benchmark that finally tests if AI agents can actually run a business for a whole year. Instead of simple chat tasks, it gives eighteen different language models a hundred thousand yen and lets them operate simulated online stores for three hundred and sixty-five days. They have to negotiate with suppliers, manage inventory, and keep their cash flow alive to maximize their final assets. The real twist is that the market is deterministic and includes fraudulent suppliers, so every mistake is the agent's own doing.

🆔 @hackernewsgithubprojects

📦 lixiaoxiao9888-create/manju-laoli-skill

Manju Laoli: The AI Director for Viral Short Dramas

manju-laoli-skill is the prompt library that turns your AI into a professional short-drama director. It stops the chaos of random text prompts by giving your AI a strict six-step workflow. Instead of guessing, it locks in your characters and locations before it even touches the script. This asset-first approach ensures every scene looks consistent and high-budget. It also includes specific rules for action scenes and dialogue pacing. It’s like handing your AI a complete film crew. No more generic outputs. Just consistent, viral-ready storytelling.

🆔 @hackernewsgithubprojects

📦 arnon-hs/open-source

Open Source Catalog

Organize thirty-four thousand open-source projects into a clear, searchable map. This repository solves the chaos of finding reliable code by automatically scanning repositories and adding practical summaries, quality signals, and transparent scores. It turns a noisy, fast-moving ecosystem into a structured decision layer, helping you compare options and spot trustworthy tools without digging through messy documentation. The catalog refreshes itself continuously, keeping up with new releases and changes. It is a living directory that removes the guesswork from choosing a library, giving you a grounded starting point for your next build. When you need a solid foundation, this organized view saves you hours of searching and guessing.

🆔 @hackernewsgithubprojects

📦 voltagent/awesome-moltbot-skills

Awesome Moltbot Skills

The Awesome Moltbot Skills list is the curated map that cuts through the noise of five thousand open source tools. It acts as a quality filter for the open source agent ecosystem, checking that every skill is actually published, tested, and safe before it makes the cut. This saves you from digging through endless, unvetted code. Instead of gambling with random scripts, you get a clean, organized library sorted into clear categories like smart home, finance, and development. It is the difference between wandering a maze and walking a straight line. Use it to find proven tools that actually work, so you can build with confidence instead of just guessing.

🆔 @hackernewsgithubprojects