📦 xhtoken/spark-x2.5

Spark-X2.5: Local AI Power

Run powerful AI agents directly on your own hardware without relying on cloud servers. Spark-X2.5 delivers serious coding and reasoning skills in a compact model that fits on personal devices. The standout feature is a unique attention design that lets the model process one million tokens of context efficiently. This means you can feed it entire codebases or long documents without slowing down or losing track. It solves the problem of needing massive, expensive data centers for everyday intelligent tasks. By making high-level AI accessible and local, it gives developers full control over their data and workflows. The result is a practical tool for anyone building smart applications locally.

📰 https://news.ycombinator.com/item?id=49907193

🆔 @hackernewsgithubprojects

📦 fastcrw/crw

crw: The Tiny Rust Web Scraper

A single Rust binary that runs in just six megabytes of memory is outpacing massive cloud services like Tavily and Firecrawl by more than twice in speed. This project is a lightweight web engine that turns any website into clean markdown or structured data for artificial intelligence agents. It solves the problem of slow, heavy scraping tools by providing a fast, local alternative that speaks a simple language your coding assistants already understand. You can run it on your own machine without a server, or use their cloud version.

🆔 @hackernewsgithubprojects

📦 mafifrizi/ares

Ares: The Automated Red Team Platform

Ares calculates the shortest path to total network compromise before a single packet is sent. This platform automates authorized red team engagements by mapping multi-stage attack paths across active directory environments, proving exploitability rather than just listing vulnerabilities. It enforces strict scope boundaries to prevent accidental out-of-scope access while storing harvested credentials in an encrypted vault. Security teams use it to validate defensive posture continuously without the cost of manual testing. The result is a clear, deterministic view of how an attacker would move through the network. This shifts security validation from guesswork to precise, auditable proof of risk.

🆔 @hackernewsgithubprojects

📦 posthog/jeeves

Jeeves: The Model That Thinks Before It Decides

Jeeves is a nine billion parameter decision model that actually pauses to think through a problem before giving you an answer. Built by PostHog, it solves a common issue where standard classifiers guess quickly but often get it wrong. By training a specific base model to reason through its logic, Jeeves achieves higher accuracy than its competitors even on new data it has never seen. It handles yes or no questions, multiple choice, and ratings all in one request while remaining fast enough for real time use. If you are tired of low accuracy in automated decision pipelines, this is a clear upgrade that proves thinking before answering pays off.

📰 https://news.ycombinator.com/item?id=49891290

🆔 @hackernewsgithubprojects

📦 arcangel0/cyberarch-shell

CyberArch Shell

CyberArch Shell turns your computer desktop into the playable interface of the video game Cyberpunk 2077. It is a custom setup for Arch Linux and the Hyprland window manager that completely replaces your standard taskbar and icons. Instead of boring menus, you get a glowing heads-up display where your battery drains like character stamina and your internet speed updates like a real-time radar. The project maps every system function to game mechanics, so checking your disk space looks like reading a cyberware status bar. If you are a developer or tech enthusiast, this repository is a fascinating look at how much control users can have over their digital environment.

🆔 @hackernewsgithubprojects

📦 duckdb/duckdb-go

DuckDB Go Driver

DuckDB Go is the official database driver that brings DuckDB's analytics power directly into your Go applications. Instead of running a separate server, this library embeds the entire database engine right inside your program, allowing you to run high-performance SQL queries on your data without the usual network overhead. It connects seamlessly with Go's standard database tools, so if you know how to write a basic query, you already know how to use this. The best part is that the database team now maintains it directly, ensuring you get the latest features and fixes without waiting on a third party.

🆔 @hackernewsgithubprojects

📦 humanbound/humanbound

Humanbound: Test your AI agents like an attacker

Break your AI agent before a real attacker does. Humanbound is an open-source testing tool that simulates hostile users by running live, multi-turn conversations against your actual software endpoint. It probes for weaknesses in how your agent uses tools and handles requests, scoring every interaction against your specific security policy. The standout feature is the automatic guardrail generation: when the tool finds a vulnerability, it immediately converts that failure into a deployable firewall rule. This means the same run that finds the hole also patches it, closing the loop between testing and defense. It is a practical way to secure your applications without relying solely on manual review.

📰 https://news.ycombinator.com/item?id=49907551

🆔 @hackernewsgithubprojects

📦 drcathicks/informed-patient

Informed Patient

Take control of your next doctor visit. Informed Patient is a tool that turns a casual chat with an A.I. into a structured investigation of your symptoms. It stops the machine from guessing and makes it search for real scientific evidence instead. You walk away with a clear list of questions to ask your doctor, backed by actual research. It helps you speak up and understand your own health. Stop guessing and start asking better questions. Your health, your voice.

🆔 @hackernewsgithubprojects

📦 fulldiagnose/antigravity-fixer

Antigravity Fixer

Antigravity Fixer is the command line tool that finally clears the confusing not eligible errors blocking access to Google's new coding assistant. When your account suddenly gets rejected, this program quietly steps in to hunt down and delete the old login tokens and cache files that are causing the trouble. It then guides you through a fresh sign in so you can get back to writing code without fighting your browser. It is a simple open source fix for a frustrating dead end, and it works on Windows, Mac, and Linux. If you are locked out, this is the quickest way to get your access back.

🆔 @hackernewsgithubprojects

📦 mizchi/explainer

Explainer: The AI That Proves It Understands

Explain complex technical concepts to a single human reader by verifying every single claim with actual code execution and visual checks. This project solves the problem where AI writes documentation faster than people can understand it, by forcing the AI to prove its points with runnable examples and diagrams that are machine-verified for accuracy. Instead of just guessing at how a system works, it builds real artifacts like state graphs and logic models, then tests them against your specific knowledge level. It is essentially a quality control system for AI-generated education.

🆔 @hackernewsgithubprojects

📦 entropyconquers/simfleet

Simfleet: The Control Plane for Parallel React Native

Simfleet is the control plane that finally keeps your parallel React Native work organized on macOS. It solves the chaos of running multiple branches at once by slimming down iOS simulators and Android emulators to save gigabytes of memory. The real magic is the shared build cache, which lets ten different coding agents test JavaScript changes simultaneously without forcing a single native rebuild. You get a live dashboard to see every device and a simple system that leases out ports so your tools never collide. It turns a messy setup of competing servers into one clean, managed fleet.

🆔 @hackernewsgithubprojects

📦 zmactep/patinae

Patinae: The GPU-Powered Molecular Viewer

Patinae brings the familiar command-line workflow of molecular visualization to a modern, graphics processing unit-first engine that runs on your desktop, in Python scripts, and directly inside a web browser. The project was built to give researchers a fast, independent tool for analyzing protein structures without the slow startup times or legacy code constraints of older viewers. It parses complex structural files and renders them with high fidelity using a flexible plugin system for AI and custom scripts. If you work with structural biology data, Patinae offers a streamlined way to explore molecular geometry with the speed of a native application and the portability of a web component.

🆔 @hackernewsgithubprojects

📦 pdparchitect/noodle

Noodle: Your AI Workspace

Noodle is the native Mac workspace that lets you manage persistent AI agents like a team. It solves the problem of messy, disconnected chat bots by giving each one a stable memory, backstory, and secure sandbox. The coolest part is that these agents can literally build and run their own mini-apps and games right on your machine, which feels like having a tiny, capable software team living inside your desktop. It is not just a chat box; it is a full environment where your bots can share files, use your local tools, and even collaborate in group chats toward shared goals.

🆔 @hackernewsgithubprojects

📦 ngwg/ceasta

Ceasta: Point an AI at any binary

Point an AI at a binary and have it debug the live process, not just read the code. Ceasta is a free, all-in-one tool that turns raw machine code into readable instructions and clean pseudocode. It handles Windows, Linux, and Mac files without needing separate setups for each. The real surprise is the built-in server that lets your AI assistant set breakpoints and step through running programs right alongside you. This makes reverse engineering far more accessible for developers who want to understand exactly what their software is doing under the hood.

🆔 @hackernewsgithubprojects

📦 rohitg00/awesome-claude-code-toolkit

135 AI Agents in One Repo

One single repository now contains one hundred and thirty five specialized artificial intelligence agents designed to handle every corner of software development. The project is called awesome claude code toolkit and it solves the problem of generic assistants lacking specific expertise. Instead of a single chatbot, you get dedicated roles like a rust systems expert or a kubernetes specialist that know exactly how to handle their specific tasks. This turns a standard coding assistant into a full virtual engineering team right in your terminal. If you want your code to be handled by specialists rather than a generalist, this collection is the place to start.

🆔 @hackernewsgithubprojects

📦 mbillow/localthings

Control Your Samsung Appliances Locally

LocalThings lets you control Samsung appliances directly from Home Assistant without touching the cloud. You type in an IP address, and it connects right over your home network, skipping the internet entirely. It handles washers, fridges, and ovens by talking to them in their own language, so your data stays private. The coolest part is that it updates instantly when something changes, giving you real-time control without waiting. If you own Samsung gear and want faster, private control, this tool is exactly what you need.

📰 https://news.ycombinator.com/item?id=49909400

🆔 @hackernewsgithubprojects

📦 espargos/esp-sdr

Turn a $3 Chip Into a Radio

Take a cheap ESP32 chip and turn it into a real software defined radio without buying extra hardware. This project, ESP-SDR, digs into undocumented debug paths inside the chip to bypass the standard Wi-Fi modem. That bypass lets the processor grab raw radio signals directly from the antenna. You can then analyze the spectrum of anything near two point four gigahertz on your computer. It is wild that you can do this with a microcontroller costing less than five dollars. No expensive spectrum analyzer required. Just a little bit of clever code and a USB cable.

📰 https://news.ycombinator.com/item?id=49893735

🆔 @hackernewsgithubprojects

📦 llmmanorg/llmman

Llmman: The Model Manager That Treats AI Like Docker

Llmman is the model manager that finally treats artificial intelligence models just like your software packages. Instead of using a messy, proprietary folder structure, it stores every model as a standard container image you already know how to handle. This means you can pull, push, and sign models using the exact same tools you use for your Docker containers, completely bypassing any specific hub or gatekeeper. It takes the complex business of moving heavy AI files and turns it into simple, standard supply chain logistics. If you want to run local models without the usual headache, this is the clean way to do it.

🆔 @hackernewsgithubprojects

📦 github/gh-actions-lock

GitHub Just Locked Your Workflows

GitHub just made it impossible for a single malicious update to hijack your entire build pipeline. The new tool called gh-actions-lock works like a digital safety seal for your automation scripts. Instead of trusting a vague version number, it records the exact, unchangeable fingerprint of every code snippet your workflows download. If anyone tries to swap in a fake or broken update, the system instantly blocks it. This turns your automated checks from a hopeful guess into a guaranteed, verified process. It is a quiet, powerful shift toward treating your build steps as critical security infrastructure rather than simple scripts.

🆔 @hackernewsgithubprojects

📦 fazlerocks/jevmail

Jevmail: The AI Inbox Sorter

Jevmail is the email triage tool that finally sorts your inbox without touching it. Instead of using a chatbot to guess, this app uses a specialized decision model to instantly categorize your Gmail messages into simple trays like Needs Reply or Spam. It is strictly read-only, meaning it can only look at your mail and never change a single label or mark anything as read. The coolest part is how it actually works: it processes a thousand emails in about a minute for just three cents, letting you see exactly which mail needs your attention while keeping everything else out of the way.

🆔 @hackernewsgithubprojects