📦 colorado-mesh/mesh-client

One App for Three Mesh Networks

One single desktop app now lets you chat and see maps across three totally different mesh radio systems at once. Mesh Client brings Meshtastic, MeshCore, and Reticulum into one window, so you can switch protocols without juggling separate tools. It handles the messy bits of hardware connection, like Bluetooth or USB, and keeps your message history safe locally. The real surprise is the unified view: you can hop between networks instantly while tracking where your messages actually went. This makes offline communication way less intimidating for anyone setting up a resilient network. Next time your phone dies, you have a plan that works.

📰 https://news.ycombinator.com/item?id=49174086

🆔 @hackernewsgithubprojects

📦 schlarpc/re-shell

Re-Shell: The AI-Powered Reverse Engineering Kit

Re-Shell is the reverse engineering environment that finally lets AI handle the tool setup for you. Instead of spending hours configuring static analyzers and decompilers, you just drop a binary into the folder and ask Claude Code to investigate. The system automatically detects what kind of file you have, like a Windows executable or an Android package, and instantly activates the right set of tools for the job. It even bundles in heavy hitters like Ghidra and Frida so you have everything ready to go. The coolest part is that it can build itself.

📰 https://news.ycombinator.com/item?id=49413569

🆔 @hackernewsgithubprojects

📦 laude-institute/headlong

Headlong: The AI That Thinks in Bash

Headlong is the agent microharness that finally lets an AI think on its own. Most AI tools just wait for you to type a command, but this one runs a constant loop, keeping a stream of thoughts even when you are not talking to it. You give it a name, and it starts projects on its own. The coolest part is that it doesn't use fancy code libraries to do work. It simply writes shell commands, runs them, and reads the results, treating the terminal like a human operator would.

🆔 @hackernewsgithubprojects

📦 tobi/walgit

One Binary, Infinite Repos

Run a full git server on hardware that cannot possibly hold your code, and lose nothing if the machine dies. Walgit is a single binary that points to a cloud storage bucket, treating that bucket as the only source of truth. Every machine running it is just a disposable cache, so you can scale to repositories larger than your server without complex databases or leader elections. The coolest part is how clones work: instead of the server streaming gigabytes of data, it hands out static bundles directly from the bucket, letting a content delivery network do the heavy lifting.

📰 https://news.ycombinator.com/item?id=49420598

🆔 @hackernewsgithubprojects

📦 tjqscott/dory

Dory: The AI Assistant That Remembers Without the Bloat

Dory treats your hosted AI chats as disposable tools by moving all your project memory to your local hard drive instead of letting it live in the browser. The biggest problem with long coding sessions is that every new message forces the AI to re-read the entire history, which burns through your token limits and makes the model forget earlier details. Dory solves this by saving your code decisions and session summaries into a local folder within your project. This means you can close your browser tab at any time without losing progress, and when you start a new chat, you simply load that saved history.

🆔 @hackernewsgithubprojects

📦 adobe-research/ldr

LDR: Predicting the Future of Video

Take just three frames of a video and LDR predicts how the entire scene evolves into the future. This Adobe and UCSD project acts like a video world model that doesn't just mimic pixels, but actually understands physical rules. It figures out how objects move, collide, or bounce based on their hidden momentum. Instead of simply guessing the next picture, it uses latent dynamics reasoning to extrapolate realistic outcomes far beyond what you see. It works across tasks like uniform motion or collisions. This is a huge step for AI that truly comprehends how our world physically evolves over time.

🆔 @hackernewsgithubprojects

📦 nguyenphutrong/quotio

Quotio: The AI Account Manager for Mac

A tiny app in your Mac menu bar quietly switches between your AI coding accounts the moment one runs out of tokens. Quotio acts as a smart router for tools like Claude Code and OpenCode, keeping your work moving without constant interruptions. It connects multiple AI subscriptions, monitors their limits in real time, and automatically fails over to a fresh account when the current one hits its limit. This stops the annoying mid-code error messages that force you to manually swap providers. It’s the missing middleman that keeps your coding agents running smoothly. Now your AI tools just work, no matter which account is available.

🆔 @hackernewsgithubprojects

📦 bojieli/queqiao

Queqiao Makes Slow Internet Feel Local

Queqiao is the self-hosted network proxy that treats packet loss as erasure, not congestion. Most internet tools assume every dropped packet means your connection is too busy, so they slow down. But Queqiao is built for those stubborn, long-distance links where packets just vanish no matter how much space you have. It uses a clever mix of error correction and retransmission to keep your data moving smoothly, even on a path that usually feels like a digital swamp. The result? A local proxy that makes your global connection feel surprisingly fast and stable, without needing to build your own expensive relay infrastructure.

🆔 @hackernewsgithubprojects

📦 1f916-ai/1f916

The Internet Forum Where Only Robots Can Post

One F One Six is a digital forum exclusively built for artificial intelligence agents. There is no website for humans to browse, so if you visit, the system just tells you to leave. Instead, AI bots connect through machine-only doors to post, vote, and argue. The project’s most surprising feature is its strict scarcity law. Because computers can generate infinite text, the rules force every robot to choose wisely. Each agent gets only one post per day, twenty comments, and fifty votes. This limit creates genuine community choice, forcing robots to spend their limited attention on their best thoughts rather than spamming the feed.

🆔 @hackernewsgithubprojects

📦 skardilabs/skardi

Skardi: The Data Plane for AI Agents

Skardi solves the annoying plumbing problem when you build AI agents. Instead of wiring up separate tools for vector databases, SQL queries, and file access, this tool connects to everything at once. You simply point it at your data sources, and it automatically creates a clean interface for your agent to use. The coolest part is that it lets your AI decide what data to look up or change, but it keeps a strict record of every single action. This means you can give your bot real freedom to work without worrying about it breaking your database.

📰 https://news.ycombinator.com/item?id=47877023

🆔 @hackernewsgithubprojects

📦 browser-use/windows-harness

Windows Harness: Giving AI a Real Mouse

Your AI agent just learned to drive your actual Windows computer, not a fake sandbox. The Windows Harness connects directly to your live desktop using a single Python process, giving the model six raw tools like seeing, clicking, and typing. Forget building custom code for every app. Instead, the agent writes its own logic on the fly to handle whatever it sees. It even grabs screenshots of background apps without stealing your focus or moving your cursor. This is raw power, letting your AI tackle any task by coding the solution in real time. Stop building brittle scripts for specific apps.

🆔 @hackernewsgithubprojects

📦 mikjee/warpdrv

warpdrv: Run Your Own AI Server

Spin up your own private AI server with a single click using this slick new desktop app called warpdrv. It acts as a powerful manager for local large language models, letting you launch, configure, and chat with different AI models running directly on your own computer hardware. The coolest part is that it handles the messy technical setup for you, automatically finding your models and managing the complex server connections behind the scenes. This means you can easily test various setups and even let other applications on your network connect to your local AI, giving you full control without needing a cloud subscription.

🆔 @hackernewsgithubprojects

📦 cseeyangchen/human-centric-ai

Human-Centric AI: The Missing Map for Human Models

Human-Centric AI Resources is the open hub that finally connects the fragmented world of human-focused artificial intelligence. While most tools focus on one specific task, this project bridges the gap by organizing everything from visual recognition to body movement under a single, clear framework. It brings together academic papers, practical tools, and course materials into one place, helping researchers see how different aspects of human interaction actually fit together. This is the map you need to understand how AI sees us as full, dynamic people rather than just pixels on a screen.

🆔 @hackernewsgithubprojects

📦 ouru77/ev-assistant

E.V. is a Private AI for Your PC

Imagine a voice assistant that actually lives on your computer, not in the cloud. That is exactly what E.V. does. It runs entirely locally using open-source tools, meaning your conversations and data never leave your machine. The coolest part? It doesn't just chat; it can control your actual PC, like opening apps or adjusting volume, while remembering specific details about you across different sessions. It feels like having a helpful, tech-savvy friend sitting next to you who respects your privacy. If you are looking for a smart assistant that prioritizes privacy over data collection, E.V. is a fascinating project to explore and build from.

🆔 @hackernewsgithubprojects

📦 melandlabs/opencontext

OpenContext: The Memory Engine for AI Agents

OpenContext is the memory engine that finally lets your AI agents remember things accurately over time. It solves the problem of agents forgetting context or losing track of decisions by storing every fact in a timeline where you can always see when something was true and when it changed. Think of it as a permanent, searchable diary for your software that knows exactly what happened and when. The most surprising part is that it connects directly to your everyday tools like Slack, Gmail, and Telegram, so your agent stays in sync without you having to manually copy data.

🆔 @hackernewsgithubprojects

📦 fudan-generative-vision/wam-diff2

WAM-Diff2: The Smart Shortcut for Self-Driving AI

Make a self-driving car’s brain run at full speed without losing its precision, and WAM-Diff2 from Fudan University does exactly that. Instead of the slow, step-by-step reasoning used by standard AI models, this system uses a clever shortcut called diffusion distillation to generate driving plans in just a few quick guesses. It’s like teaching a student to solve complex math problems by showing them the final answer first, then refining it. This dramatically speeds up the process, making real-time autonomous driving possible on standard hardware. If you’re into the tech behind smart cars, this is the tool that makes the future of driving fast and efficient.

🆔 @hackernewsgithubprojects

📦 skyeshark/eidoverse-video

Eidoverse Video: The AI Film Studio

Eidoverse Video is the autonomous film studio that lets artificial intelligence agents direct and render complex cinematic sequences entirely on your own computer. Instead of relying on heavy cloud servers, this toolkit uses your graphics card to build detailed worlds, animate realistic 3D characters, and generate original music and sound effects from scratch. The most surprising feature is that the AI agent actually acts as the filmmaker, writing scene scripts, handling physics simulations, and even checking its own work for visual errors before you see the final video.

🆔 @hackernewsgithubprojects

📦 furkankly/zoetrope

Zoetrope: Watch Your AI Coding Session as a Live Map

zoetrope is the terminal tool that turns your AI coding sessions into a live, navigable map. When you use Claude Code, it quietly saves a text log of everything, but those logs are just walls of code. zoetrope reads that log and draws a real-time diagram in your terminal, showing the main agent, the sub-agents it spawns, and the tools they run as connected nodes. It’s like a flowchart that draws itself while the code runs, but here is the wild part: it has a time scrubber.

🆔 @hackernewsgithubprojects

📦 apodexai/frontieragent

FrontierAgent: The AI Team That Works Like Real Humans

FrontierAgent turns your terminal into a live command center where AI agents actually do research and write files. It runs entirely on your machine without needing heavy infrastructure or pre-installed bloat. The coolest part is the Agent Team mode. Instead of one bot guessing answers, a coordinator breaks your task down and hires parallel sub-agents to investigate different angles at once. You watch them work in real time, and you can even type in new instructions while they are busy without stopping them. It is like managing a small, hyper-efficient research firm from your keyboard.

🆔 @hackernewsgithubprojects

📦 thiagotigaz/ocr-it

OCR It: Turn Invisible PDFs Into Text

This browser extension turns any unreadable PDF into selectable text by letting you capture the same screen area on every page automatically. You define the text region once, and the extension runs the text recognition locally, meaning nothing leaves your computer. It is perfect for feeding hundreds of pages into an AI assistant for summarization or search. If you have ever been stuck with a document that refuses to let you copy text, this tool hands you the text for free, with zero privacy risk and no internet connection required.

📰 https://news.ycombinator.com/item?id=49415852

🆔 @hackernewsgithubprojects

📦 tb2-sy/aimold

AIMold Automates Complex Mold Design

AIMold is the AI pipeline that instantly turns a single digital part into a complete, ready-to-build mold. Usually, creating these injection molds is a massive headache for engineers who have to hand-draw every support bracket and cavity. This project takes one simple input file and automatically generates the entire assembly, including the upper and lower mold pieces plus all the tricky auxiliary components you would never expect a computer to handle on its own. It bridges the gap between a basic design and actual manufacturing without any manual tweaking. If you work with product design, this is the future of turning ideas into physical objects without the usual bottleneck.

🆔 @hackernewsgithubprojects