📦 toicf/cf-workers-tgproxy

CF-Workers-TGProxy

Route your desktop or Android Telegram traffic directly to the official server using just one file. This project removes the heavy middleman server you usually need to run a private Telegram proxy. Instead of trusting a third party to decrypt your messages, it runs on Cloudflare Workers to simply swap the transport layer wrapper. Your actual chat content stays fully encrypted between your device and Telegram the whole time. The worker just peels off the outer shell, adds a fresh random shell, and forwards it to the official endpoint. It is a clever way to get fast, secure connections without maintaining your own dedicated infrastructure.

🆔 @hackernewsgithubprojects

📦 helixdb/helix-foundry

Helix Foundry: Your Data, Your Rules

Helix Foundry is the local-first data workspace that connects all your company's scattered tools into one unified map right on your laptop. Instead of trusting a cloud vendor with your sensitive customer records, this tool pulls data from your databases, billing systems, and file servers to build a single, connected view you can actually explore. The coolest part is that it runs entirely offline by default, meaning your data never leaves your machine, and you can ask questions about your business in plain English. It is essentially a private data engine that gives you full ownership without the usual security headaches.

🆔 @hackernewsgithubprojects

📦 mosi-ai/roboicl

RoboICL

RoboICL lets a frozen, general-purpose language model control a physical robot arm without ever training it. It works by showing the model a few recorded examples of a task, like depositing a coin, and then letting it watch its own live attempts to correct mistakes in real time. The system feeds camera feeds and sensor data directly into the model, which predicts precise movements for the robot to execute. This means you can teach a new task to a large AI model just by providing a few demos, no custom robot brain required. It turns a generic AI into a capable robot learner on the fly.

🆔 @hackernewsgithubprojects

📦 datagouv/datagouv-mcp

Ask your AI about French government data

Stop scrolling through endless government websites to find that one specific CSV file. DataGouv MCP lets you simply ask your favorite AI chatbot for information about French open data, and it finds the answer instantly. Instead of manually browsing the catalog for housing prices or population stats, you just type your question in plain language, and the tool digs through the official national platform to pull up the exact datasets you need. It connects any compatible chatbot directly to the source, turning a tedious search into a quick conversation. No more guessing file names or hunting through folders; you get straight to the data.

🆔 @hackernewsgithubprojects

📦 goodnight000/api-anything

API Anything: Turn Websites into APIs

Teach your AI agent a website just once, and it instantly learns to call that site like a standard software interface instead of slowly clicking through a browser. This tool records the hidden data requests a page makes, extracts the useful parts, and saves them as reusable functions. The coolest part is that it fixes itself. If a website changes its layout or breaks the data structure, the tool automatically repairs the connection so your agent never breaks. You stop paying for slow, expensive browser automation and start getting fast, clean data instantly. It is the perfect bridge between messy websites and reliable code.

📰 https://news.ycombinator.com/item?id=50017682

🆔 @hackernewsgithubprojects

📦 anthropics/oss-scanner

Oss Scanner

Get your critical open-source code professionally scanned for security vulnerabilities, free of charge, directly from Anthropic. This project, called Oss Scanner, solves a major problem for maintainers who need expert security audits but lack the budget. The coolest part? You enroll your project by simply opening a pull request in this repository. Once merged, Anthropic builds your code in an isolated virtual machine, completely cuts off the internet, and runs a deep security audit. They then email you the findings, complete with reproduction steps and proposed patches. It is private, non-public, and designed specifically for the open-source ecosystem. No public disclosure periods, just private fixes.

🆔 @hackernewsgithubprojects

📦 aipentest/cyberstrikeai

CyberStrikeAI: The AI Pentest Workspace

CyberStrikeAI runs an entire security audit from a single sentence, turning a simple request into a governed, step-by-step hacking plan that requires your explicit approval before touching anything. It orchestrates a team of specialized A.I. agents to map attack surfaces, test for vulnerabilities, and manage the evidence, all within one auditable workspace. The most fascinating part is how it treats every action as a lesson; the system remembers what worked and what failed, so it gets smarter with every operation. It is not just a scanner, but a full operational memory for authorized testing.

🆔 @hackernewsgithubprojects

📦 kvmem/kvmem-llama.cpp

KVMem: Run 256k Context on a 16GB GPU

KVMem is the memory manager that fits massive two hundred and fifty six thousand token contexts onto a sixteen gigabyte graphics card. Running long-running agents usually chokes your memory because the attention cache grows endlessly, but this tool solves that by moving old conversation history to your system RAM and only fetching the relevant bits back when needed. It is not just about saving space, though. The paper proves that keeping just thirty two thousand tokens active on the GPU is nearly lossless compared to holding the whole history, meaning your agent stays smart without the slowdown.

🆔 @hackernewsgithubprojects

📦 xiaozhou26/chromix

Chromix: The Browser That Remembers You

Chromix is a custom-built browser that lets you script exactly what it looks like to the internet. Unlike standard automation tools that often get flagged as bots, this project modifies the browser source code to create a consistent digital persona. You can set the language, timezone, screen size, and even the graphics card model all from one simple configuration file. The coolest part is that it remembers this identity across sessions, so your browser doesn't just look human; it acts like the same specific user every time. It works with both Python and JavaScript, making it incredibly flexible for developers who need reproducible tests.

🆔 @hackernewsgithubprojects

📦 tencent-hunyuan/hy-mt2

Tiny Translator: Tencent's New Model

Translate text between thirty-three languages using a model so small it fits on your phone. Tencent released Hy-MT2, a family of translation models that swap out giant servers for lightweight, on-device power. The biggest surprise is the 1.8 billion parameter version, which shrinks down to just four hundred forty megabytes through aggressive compression. That means you can run high-quality, instruction-following translation offline, without needing a cloud connection or a massive GPU. It is not just a dictionary; it understands context, preserves code structure, and adapts to specific styles. This is what real-world translation looks like when speed and privacy matter more than raw size.

🆔 @hackernewsgithubprojects

📦 liguobao/ds-harness-remote

DeepSeek Harness Remote: Access Your AI From Anywhere

Pick up your work from your phone without closing your laptop. DeepSeek Harness Remote lets you connect to the AI coding environment running on your desktop computer using your phone, tablet, or a web browser. This solves a major frustration for developers who want to monitor or adjust their AI agent sessions while away from their desk. The standout feature is its ability to work securely through any internet connection without needing to expose your computer to the public web. It creates a private, encrypted bridge between your devices, ensuring your code and prompts stay safe.

🆔 @hackernewsgithubprojects

📦 agent-field/codeaf

CodeAF: The Factory for Open AI

CodeAF is an open-source software factory built specifically to manage and benchmark open models. It solves the problem of chaotic agent workflows by providing a structured environment where you can test, route, and observe how different AI models handle real-world coding tasks. Instead of just chatting with an AI, you get a rigorous lab to see exactly where it succeeds or fails on specific jobs. This makes it incredibly useful for developers who need reliable, open-source intelligence without the black box. It is less about writing code and more about building a trustworthy system around it.

🆔 @hackernewsgithubprojects

📦 pimpinpumpkin/vela

Vela: The Google Maps Alternative

Navigate anywhere with full turn-by-turn directions, live traffic, and rich place details, all on a phone stripped of every Google service. This app is built for privacy-first Android users who want the familiar feel of Google Maps without the account, the Play Services dependency, or the constant data pings. It renders maps using open source data, so you can browse streets and search locations without anyone watching. When you actually need directions, it cleverly uses Google only for the route calculation, keeping the rest of your journey private. It is a native app that works offline, meaning you can drive through a dead zone and still get guidance.

🆔 @hackernewsgithubprojects

📦 hinln/artex

How Artex Automates Security Testing

Artex is the autonomous penetration testing system that turns complex security audits into a visual, collaborative workflow. It replaces manual guessing with a smart AI planner that maps out specific attack paths, while a team of digital workers executes those steps in real-time. The coolest part is the live exploration graph, which visually connects every test result to the specific asset being scanned, so you can instantly see exactly where your coverage gaps are. It even lets human security experts step in to approve sensitive actions, keeping you in control.

🆔 @hackernewsgithubprojects

📦 bincequ/roboharness

RoboHarness: AI Robots That See Like You

RoboHarness is the robotic control system that lets AI agents physically interact with the world using simple visual cues. Instead of training massive, expensive models to learn how to pick up a cup from scratch, this project gives an existing language model a direct line of sight to a robot's camera. The AI just marks a dot on the image where it wants to grab, and the system handles the heavy math to translate that two-dimensional click into the precise three-dimensional movement needed to execute the task.

🆔 @hackernewsgithubprojects

📦 dayuanjiang/pecofence

PecoFence: Let Your AI Organize Your Windows 11 Desktop

PecoFence is a free, open-source desktop organizer that lets your favorite AI agent manage your Windows 11 files using a built-in command line tool. It creates glass-like panels to hold your work, automatically sorting new files into the right places based on rules you define. The standout feature is its ability for tools like Claude to read your current setup and apply changes directly, such as adjusting transparency or creating new project folders. You can even save snapshots to restore your layout later if things get messy. This open-source alternative to paid software is built in Rust, supports ten languages, and runs entirely locally without needing an account.

🆔 @hackernewsgithubprojects

📦 antfu/skills-pack

skills-pack

Give your AI coding assistants a serious education with a single command. This tool bundles a carefully selected collection of instructions, or skills, from various open-source projects into one installable package. Instead of hunting down individual guides for code review, testing, or writing clear documentation, you install this pack and let it automatically wire those capabilities into your local development environment. The standout feature is how it links these external skills to the AI agents you already use, like Claude Code or Cursor. This means your tools instantly gain specialized expertise for specific tasks, such as rigorous code reviews or structured research, without you needing to manually manage configuration files.

🆔 @hackernewsgithubprojects

📦 parachord/parachord

Parachord: Your Music, Everywhere

Parachord plays your music from everywhere at once. It finds the best source for every track, whether that’s Spotify, YouTube, Bandcamp, or files on your hard drive. No more switching apps. Just one interface. It even lets you drag and drop to reorder which service you prefer. That means you can support indie artists on Bandcamp while still falling back to YouTube for rare tracks. All in one flow. It’s a smart, flexible way to listen to everything you own. Your music is scattered. Parachord brings it together. Give it a try. It’s surprisingly good.

🆔 @hackernewsgithubprojects

📦 rankftw/rhi

RHI: One-Click HDR for Every Game

RHI is the game library manager that finally makes HDR setup painless. It scans your entire PC collection across every major store and handles the messy, technical side of visual enhancements automatically. You get one-click installation for ReShade, HDR mods, and driver profiles, meaning you skip the tedious file renaming and manual config editing that usually breaks immersion. It is the ultimate time saver for players who want their games to look incredible without becoming an IT support ticket for their own computer. If you play on Windows, this is the shortcut you have been waiting for.

🆔 @hackernewsgithubprojects

📦 mrlesk/groma.md

Groma: Your Code as a City

Groma turns your entire software architecture into a living, three-dimensional city map that lives right inside your code repository. Instead of static diagrams that rot the moment you change a file, this tool scans your actual source code and renders your systems as buildings and streets. You can walk through this digital landscape, zooming in from a high-level overview all the way down to individual files. It solves the problem of architecture drift by keeping your visual map in sync with your code.

🆔 @hackernewsgithubprojects

📦 yuliang-liu/awesome-ocr-llm

Awesome OCR LLM: The Future of Document Reading

Awesome OCR LLM is the curated collection that maps the rapid shift from rule-based text extraction to unified visual intelligence. It solves the problem of fragmented research by tracking how modern vision-language models now read entire documents in one go, handling complex layouts without needing a separate step for every image component. This resource is vital for anyone following how computers move past simple character recognition to truly understanding the structure and meaning of scanned pages, financial reports, and handwritten notes. If you are curious about the technology behind reading machines, this single list is your best map to the current state of the art.

🆔 @hackernewsgithubprojects