📦 lantunnel/lantunnel

Lantunnel: Your Private Network Anywhere

Connect your home and office computers into one private mesh. Lantunnel lets you reach devices on different networks directly, without opening ports or exposing anything to the internet. It works by having your devices find each other and talk face-to-face whenever possible. If the network blocks that direct link, it falls back to an encrypted relay that forwards data it cannot read. This keeps your traffic secure and private, even across different locations. Whether you are streaming media, accessing a server, or just checking a printer, everything stays inside your own secure bubble. It is a clever, lightweight way to own your connectivity.

📰 https://news.ycombinator.com/item?id=49593653

🆔 @hackernewsgithubprojects

📦 maxhu-xuan/chat-archive-guard

Chat Archive Guard: The Local Privacy Shield

Chat Archive Guard never sends your data to the cloud to check for leaks, keeping every audit entirely on your own device. It acts as a strict, read-only gate for your chat exports, scanning files locally to spot hidden passwords, personal details, and broken code structures before you share them. By refusing to upload your archive, it ensures your private conversations stay private while still giving you a clear report on what might be exposed. If you handle sensitive chat logs, this tool gives you the peace of mind to move data safely without risking a data breach, making it a simple but powerful addition to your privacy toolkit.

🆔 @hackernewsgithubprojects

📦 egma-ai/egma

Test Your Voice Agents with Egma

Egma lets you rehearse your voice agents in a simulated phone call before you ever let them talk to a real customer. You build a test suite that mocks out the tools your agent uses, so you can verify it handles edge cases like angry callers or dropped calls without risking a bad production experience. It runs these simulations right inside your code repository, giving you a safety net to catch bugs before they happen. The best part is the autonomy; you can bring your own API keys to keep costs low, and the platform is designed to be self-hosted.

📰 https://news.ycombinator.com/item?id=49646928

🆔 @hackernewsgithubprojects

📦 maxhu-xuan/task-state-guard

TaskStateGuard: The Anti-Guessing Safety Net

TaskStateGuard refuses to guess that your AI agent succeeded just because the code stopped moving. After a service restarts, it acts as a strict accountant for your background jobs, looking at a local SQLite database to find tasks stuck in a running state. It waits for real proof of success or a confirmed delivery before closing the loop, rather than just assuming everything worked out. If a task hits its deadline without that proof, it marks it as timed out, keeping your system honest. You can preview these fixes in a dry run mode to see exactly what changes before applying them.

🆔 @hackernewsgithubprojects

📦 f0rr0/oliphaunt

Oliphaunt: The Database That Fits In Your Pocket

Oliphaunt runs a full instance of Postgres directly inside your application code. No Docker containers, no external servers, and no network latency. It behaves exactly like a simple file-based database, making it perfect for mobile apps and unit tests. You get the power of a professional relational engine without the heavy infrastructure usually required. This project solves the problem of needing complex data handling in environments that cannot support a traditional server. It allows developers to use advanced SQL features locally with minimal setup. The result is a database experience that feels instant and private. Your data stays exactly where you put it.

📰 https://news.ycombinator.com/item?id=47918882

🆔 @hackernewsgithubprojects

📦 elixir-dux/dux

Dux: The Elixir DataFrame That Actually Scales

Dux is the data tool that finally makes distributed computing in Elixir feel effortless and native. You write your familiar chain of filters and summaries, but behind the scenes, the whole thing compiles into optimized SQL that runs directly on a high-speed engine called DuckDB. The coolest part is how it handles scale: just add one line to your code, and Dux automatically splits your data across multiple servers, letting them crunch numbers in parallel without any complex setup. It is fast, it is clean, and it keeps your data right where it lives.

📰 https://news.ycombinator.com/item?id=47594412

🆔 @hackernewsgithubprojects

📦 unity-technologies/unity-agent-plugin

Unity Agent Plugin: Your AI Coding Sidekick for Unity

Unity Agent Plugin is the bridge that gives your AI coding assistant the specialized skills needed to build games. Instead of generic advice, it feeds your agent specific, documented instructions for handling tricky Unity tasks like pixel-perfect rendering or setting up in-app purchases. This means when you ask your assistant to fix blurry sprites or configure multiplayer voice chat, it knows exactly which settings to tweak to get the job done right. You stop guessing and start shipping, turning your AI tool into a reliable partner that understands the unique quirks of the game engine you are working with every day.

🆔 @hackernewsgithubprojects

📦 guqiong96/lsglang

lsglang

Run massive AI models on hardware that looks too small to hold them. Lsglang takes a standard language model server and teaches it to split the work between your graphics card and your regular computer memory. Instead of cramming everything into the limited graphics card, it keeps the heavy parts in standard system memory and only moves what is needed to the fast chip. This means a single machine can handle a model that is effectively twice as large as the memory card alone allows. It is a clever way to squeeze more performance out of existing hardware without buying a new server.

🆔 @hackernewsgithubprojects

📦 uiuing/browser-agent

Browser Agent

Browser Agent is the open-source browser runtime that verifies every single action against the live page structure. It acts as a Chrome extension where the AI plans its next move in a side panel, but the real work happens on actual web pages. The project solves the problem of AI hallucination by checking if a button was truly clicked or if a list actually updated, rather than just trusting the model’s word. This makes it a reliable tool for automating real-world tasks safely. If you want an AI that proves it worked, this is the one to watch.

🆔 @hackernewsgithubprojects

📦 ivanfioravanti/ds4-metal

Run Massive AI Models Locally

Run powerful language models directly on your personal hardware with DwarfStar, a specialized inference engine built for Apple Silicon. It executes the DeepSeek V4 Flash model using your Mac’s dedicated graphics processor, eliminating the need for expensive cloud servers. This project solves the problem of running high-quality, large-scale AI locally by providing a native, optimized tool that works out of the box. You get a self-contained system that handles everything from model loading to interactive chat, making advanced AI capabilities accessible on consumer laptops. This dedicated approach ensures stable performance without the overhead of general-purpose runners.

🆔 @hackernewsgithubprojects

📦 katiahayati/lucasartsifier

How to Fix Old Adventure Game Bugs

Lucas Artsifier hunts down hidden traps in thirty-year-old adventure games where you can get permanently stuck even though the game still accepts your inputs. It reads the game's internal code, figures out exactly which items or actions cause a dead end, and automatically writes the fix right back into the game files. The coolest part is that it finds these bugs without you telling it where the goal is or how to win; it discovers everything from the code itself. It even checks its own fixes to make sure it didn't break anything else, then repacks the game so it plays perfectly.

📰 https://news.ycombinator.com/item?id=49355607

🆔 @hackernewsgithubprojects

📦 pgrundev/pgterm

pgterm: The Postgres Monitor You Need

Keep every Postgres database healthy without juggling separate windows or risking accidental writes. pgterm is a terminal interface that acts as a central dashboard for all your database environments. It runs read-only diagnostics in the background to spot issues like blocked locks or unused indexes before they become problems. The standout feature is how it handles security; it strictly refuses any write commands and keeps connection details in memory, so your passwords never touch the disk. It even flags specific databases that need attention while you focus on others, keeping your workflow smooth. If you manage Postgres, this tool gives you a clear, safe view of your infrastructure.

📰 https://news.ycombinator.com/item?id=49649155

🆔 @hackernewsgithubprojects

📦 wassermanproductions/scriptbreak

ScriptBreak: The Screenplay App That Doesn't Phone Home

ScriptBreak is the desktop app that finally keeps your script private while making it AI-ready. It solves the problem of needing to upload sensitive creative work to a cloud server just to get analysis. You simply drop in your screenplay, and the app breaks it down into scenes, characters, and shot lists entirely on your machine. The most surprising part is how it handles AI. Instead of using API keys, it exports self-contained prompt files. These files carry all your production data and instructions, so you can paste them into any local or cloud-based model you already trust. No servers, no accounts, and no data leaves your computer.

🆔 @hackernewsgithubprojects

📦 ksimback/hermes-ecosystem

Hermes Atlas: The 240 Tool Decision Layer

Hermes Atlas is the decision layer that turns a sprawling AI agent ecosystem into a clear, actionable plan. It solves the problem of analysis paralysis by curating over two hundred open-source tools, presenting live GitHub star counts, and recommending the exact installation modes and skills you need for your specific workflow. Instead of guessing which plugins fit your setup, it analyzes your use case and matches it against the ecosystem data. This project is useful because it acts as a trusted guide, filtering out noise and highlighting the most relevant community projects. It transforms a chaotic landscape of options into a structured path for developers.

🆔 @hackernewsgithubprojects

📦 luchasarie/bro-skill

Bro Skill: The AI Translator for Tech Jargon

A single markdown file now forces AI assistants to stop sounding like robots and start sounding like smart friends. The project called Bro Skill solves a very specific frustration: when your coding assistant gives you a reply packed with heavy jargon that makes you want to scream, you simply type the command slash bro. The tool does not answer a new question or change the facts. Instead, it takes its own previous message and re-explains it in plain, everyday language. It keeps every file path, command, and number exactly the same, but wraps the meaning in a casual, direct tone.

📰 https://news.ycombinator.com/item?id=49649589

🆔 @hackernewsgithubprojects

📦 phenx/trazor

Trazor: The Browser-Based Vector Tracer

Turn any flat photo into crisp, editable vector art without sending a single pixel to a server. Trazor is a free, open source tool that runs entirely in your web browser to convert standard images into scalable graphics. While most online converters just guess at shapes, Trazor calculates exact mathematical curves for every color layer separately. This means the lines are perfectly smooth and the colors stay exactly where you want them. Because it works locally on your computer, your private designs never leave your device. It is a clean, precise way to get professional quality vectors from simple pictures, all from your desktop.

📰 https://news.ycombinator.com/item?id=49410743

🆔 @hackernewsgithubprojects

📦 danieljvdm/effect-agent

Effect Agent: The TypeScript Agent That Actually Works

Effect Agent is the TypeScript framework that finally makes your AI agents behave like reliable software instead of unpredictable scripts. You define exactly what goes in and what must come out, and it validates the result against your rules before you ever see it. It handles the messy parts of running tools, manages streaming updates, and keeps everything typed so you know exactly where things stand. This means your agents can plan trips, browse the web, or review code without crashing or hallucinating data. If you are building serious bots in JavaScript, this is the structure you have been missing.

📰 https://news.ycombinator.com/item?id=49580141

🆔 @hackernewsgithubprojects

📦 nvlabs/sol-pi

SoL-Pi Makes Coding Agents Smarter and Cheaper

Sol-Pi is the coding agent extension that makes your AI tools cheaper without making them dumber. It fixes the annoying habit of coding bots where they re-read huge logs and repeat the same steps over and over. The coolest part is how it bundles a file edit and its check into one single move, so the computer doesn't have to pause and think again. It also squeezes messy data into neat summaries so your agent stops wasting money on useless context. It plugs right into your existing setup without changing anything underneath. Basically, it helps your AI work smarter and save you cash.

🆔 @hackernewsgithubprojects

📦 nvlabs/enpire

ENPIRE: The Robot That Codes Its Own Brain

NVIDIA's ENPIRE lets an AI coding agent physically test its own robot code to fix errors. Usually, engineers write logic and hope it works in the real world. Here, the software agent actually drives the robot, watches it fail, and rewrites its own instructions to try again. It repeats this physical trial and error cycle until the task succeeds. This loop allows robots to learn from reality instead of just simulation. You get a system that improves itself through tangible physical proof rather than digital guesses. If you want to know how agents handle real-world hardware, this is the project to watch.

🆔 @hackernewsgithubprojects

📦 miaai-lab/qwen3.8-27b-16gb-nvidia-gpus-one-click-install

Run a 27B AI on One GPU

Run a massive twenty-seven billion parameter language model on a single consumer graphics card. This kit takes a huge artificial intelligence brain and squeezes it into your existing hardware, so you get enterprise-level intelligence without buying a data center. It automatically detects your specific card, downloads the right compressed version of the model, and sets up everything you need to start chatting immediately. You get a clean interface to talk to the AI and a standard connection point for your other favorite tools. It handles the complicated technical setup behind the scenes, meaning you just double-click and go. It works on both Windows and Linux with the same easy experience.

🆔 @hackernewsgithubprojects